Skip to content

GitLab

Commit f969ca34 authored by dongchy920's avatar dongchy920
Browse files

yolov9_pytorch

parents
Pipeline #999 failed with stages
in 0 seconds
Show whitespace changes
Inline Side-by-side
Showing with 8298 additions and 0 deletions
models/segment/yolov9-c-dseg.yaml 0 → 100644
View file @ f969ca34
# YOLOv9
# parameters
nc: 80 # number of classes
depth_multiple: 1.0 # model depth multiple
width_multiple: 1.0 # layer channel multiple
#activation: nn.LeakyReLU(0.1)
#activation: nn.ReLU()
# anchors
anchors: 3
# gelan backbone
backbone:
[
[-1, 1, Silence, []],
# conv down
[-1, 1, Conv, [64, 3, 2]], # 1-P1/2
# conv down
[-1, 1, Conv, [128, 3, 2]], # 2-P2/4
# elan-1 block
[-1, 1, RepNCSPELAN4, [256, 128, 64, 1]], # 3
# avg-conv down
[-1, 1, ADown, [256]], # 4-P3/8
# elan-2 block
[-1, 1, RepNCSPELAN4, [512, 256, 128, 1]], # 5
# avg-conv down
[-1, 1, ADown, [512]], # 6-P4/16
# elan-2 block
[-1, 1, RepNCSPELAN4, [512, 512, 256, 1]], # 7
# avg-conv down
[-1, 1, ADown, [512]], # 8-P5/32
# elan-2 block
[-1, 1, RepNCSPELAN4, [512, 512, 256, 1]], # 9
]
# YOLOv9 head
head:
[
# elan-spp block
[-1, 1, SPPELAN, [512, 256]], # 10
# up-concat merge
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
[[-1, 7], 1, Concat, [1]], # cat backbone P4
# elan-2 block
[-1, 1, RepNCSPELAN4, [512, 512, 256, 1]], # 13
# up-concat merge
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
[[-1, 5], 1, Concat, [1]], # cat backbone P3
# elan-2 block
[-1, 1, RepNCSPELAN4, [256, 256, 128, 1]], # 16 (P3/8-small)
# avg-conv-down merge
[-1, 1, ADown, [256]],
[[-1, 13], 1, Concat, [1]], # cat head P4
# elan-2 block
[-1, 1, RepNCSPELAN4, [512, 512, 256, 1]], # 19 (P4/16-medium)
# avg-conv-down merge
[-1, 1, ADown, [512]],
[[-1, 10], 1, Concat, [1]], # cat head P5
# elan-2 block
[-1, 1, RepNCSPELAN4, [512, 512, 256, 1]], # 22 (P5/32-large)
# multi-level reversible auxiliary branch
# routing
[5, 1, CBLinear, [[256]]], # 23
[7, 1, CBLinear, [[256, 512]]], # 24
[9, 1, CBLinear, [[256, 512, 512]]], # 25
# conv down
[0, 1, Conv, [64, 3, 2]], # 26-P1/2
# conv down
[-1, 1, Conv, [128, 3, 2]], # 27-P2/4
# elan-1 block
[-1, 1, RepNCSPELAN4, [256, 128, 64, 1]], # 28
# avg-conv down fuse
[-1, 1, ADown, [256]], # 29-P3/8
[[23, 24, 25, -1], 1, CBFuse, [[0, 0, 0]]], # 30
# elan-2 block
[-1, 1, RepNCSPELAN4, [512, 256, 128, 1]], # 31
# avg-conv down fuse
[-1, 1, ADown, [512]], # 32-P4/16
[[24, 25, -1], 1, CBFuse, [[1, 1]]], # 33
# elan-2 block
[-1, 1, RepNCSPELAN4, [512, 512, 256, 1]], # 34
# avg-conv down fuse
[-1, 1, ADown, [512]], # 35-P5/32
[[25, -1], 1, CBFuse, [[2]]], # 36
# elan-2 block
[-1, 1, RepNCSPELAN4, [512, 512, 256, 1]], # 37
[31, 1, RepNCSPELAN4, [512, 256, 128, 2]], # 38
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
[-1, 1, Conv, [256, 3, 1]], # 40
[16, 1, RepNCSPELAN4, [256, 256, 128, 2]], # 41
[-1, 1, nn.Upsample, [None, 2, 'nearest']],
[-1, 1, Conv, [256, 3, 1]], # 43
# segment
[[31, 34, 37, 16, 19, 22, 40, 43], 1, DualDSegment, [nc, 32, 256]], # Segment(P3, P4, P5)
]
models/tf.py 0 → 100644
View file @ f969ca34
import argparse
import sys
from copy import deepcopy
from pathlib import Path
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
# ROOT = ROOT.relative_to(Path.cwd()) # relative
import numpy as np
import tensorflow as tf
import torch
import torch.nn as nn
from tensorflow import keras
from models.common import (C3, SPP, SPPF, Bottleneck, BottleneckCSP, C3x, Concat, Conv, CrossConv, DWConv,
DWConvTranspose2d, Focus, autopad)
from models.experimental import MixConv2d, attempt_load
from models.yolo import Detect, Segment
from utils.activations import SiLU
from utils.general import LOGGER, make_divisible, print_args
class TFBN(keras.layers.Layer):
# TensorFlow BatchNormalization wrapper
def __init__(self, w=None):
super().__init__()
self.bn = keras.layers.BatchNormalization(
beta_initializer=keras.initializers.Constant(w.bias.numpy()),
gamma_initializer=keras.initializers.Constant(w.weight.numpy()),
moving_mean_initializer=keras.initializers.Constant(w.running_mean.numpy()),
moving_variance_initializer=keras.initializers.Constant(w.running_var.numpy()),
epsilon=w.eps)
def call(self, inputs):
return self.bn(inputs)
class TFPad(keras.layers.Layer):
# Pad inputs in spatial dimensions 1 and 2
def __init__(self, pad):
super().__init__()
if isinstance(pad, int):
self.pad = tf.constant([[0, 0], [pad, pad], [pad, pad], [0, 0]])
else: # tuple/list
self.pad = tf.constant([[0, 0], [pad[0], pad[0]], [pad[1], pad[1]], [0, 0]])
def call(self, inputs):
return tf.pad(inputs, self.pad, mode='constant', constant_values=0)
class TFConv(keras.layers.Layer):
# Standard convolution
def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True, w=None):
# ch_in, ch_out, weights, kernel, stride, padding, groups
super().__init__()
assert g == 1, "TF v2.2 Conv2D does not support 'groups' argument"
# TensorFlow convolution padding is inconsistent with PyTorch (e.g. k=3 s=2 'SAME' padding)
# see https://stackoverflow.com/questions/52975843/comparing-conv2d-with-padding-between-tensorflow-and-pytorch
conv = keras.layers.Conv2D(
filters=c2,
kernel_size=k,
strides=s,
padding='SAME' if s == 1 else 'VALID',
use_bias=not hasattr(w, 'bn'),
kernel_initializer=keras.initializers.Constant(w.conv.weight.permute(2, 3, 1, 0).numpy()),
bias_initializer='zeros' if hasattr(w, 'bn') else keras.initializers.Constant(w.conv.bias.numpy()))
self.conv = conv if s == 1 else keras.Sequential([TFPad(autopad(k, p)), conv])
self.bn = TFBN(w.bn) if hasattr(w, 'bn') else tf.identity
self.act = activations(w.act) if act else tf.identity
def call(self, inputs):
return self.act(self.bn(self.conv(inputs)))
class TFDWConv(keras.layers.Layer):
# Depthwise convolution
def __init__(self, c1, c2, k=1, s=1, p=None, act=True, w=None):
# ch_in, ch_out, weights, kernel, stride, padding, groups
super().__init__()
assert c2 % c1 == 0, f'TFDWConv() output={c2} must be a multiple of input={c1} channels'
conv = keras.layers.DepthwiseConv2D(
kernel_size=k,
depth_multiplier=c2 // c1,
strides=s,
padding='SAME' if s == 1 else 'VALID',
use_bias=not hasattr(w, 'bn'),
depthwise_initializer=keras.initializers.Constant(w.conv.weight.permute(2, 3, 1, 0).numpy()),
bias_initializer='zeros' if hasattr(w, 'bn') else keras.initializers.Constant(w.conv.bias.numpy()))
self.conv = conv if s == 1 else keras.Sequential([TFPad(autopad(k, p)), conv])
self.bn = TFBN(w.bn) if hasattr(w, 'bn') else tf.identity
self.act = activations(w.act) if act else tf.identity
def call(self, inputs):
return self.act(self.bn(self.conv(inputs)))
class TFDWConvTranspose2d(keras.layers.Layer):
# Depthwise ConvTranspose2d
def __init__(self, c1, c2, k=1, s=1, p1=0, p2=0, w=None):
# ch_in, ch_out, weights, kernel, stride, padding, groups
super().__init__()
assert c1 == c2, f'TFDWConv() output={c2} must be equal to input={c1} channels'
assert k == 4 and p1 == 1, 'TFDWConv() only valid for k=4 and p1=1'
weight, bias = w.weight.permute(2, 3, 1, 0).numpy(), w.bias.numpy()
self.c1 = c1
self.conv = [
keras.layers.Conv2DTranspose(filters=1,
kernel_size=k,
strides=s,
padding='VALID',
output_padding=p2,
use_bias=True,
kernel_initializer=keras.initializers.Constant(weight[..., i:i + 1]),
bias_initializer=keras.initializers.Constant(bias[i])) for i in range(c1)]
def call(self, inputs):
return tf.concat([m(x) for m, x in zip(self.conv, tf.split(inputs, self.c1, 3))], 3)[:, 1:-1, 1:-1]
class TFFocus(keras.layers.Layer):
# Focus wh information into c-space
def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True, w=None):
# ch_in, ch_out, kernel, stride, padding, groups
super().__init__()
self.conv = TFConv(c1 * 4, c2, k, s, p, g, act, w.conv)
def call(self, inputs): # x(b,w,h,c) -> y(b,w/2,h/2,4c)
# inputs = inputs / 255 # normalize 0-255 to 0-1
inputs = [inputs[:, ::2, ::2, :], inputs[:, 1::2, ::2, :], inputs[:, ::2, 1::2, :], inputs[:, 1::2, 1::2, :]]
return self.conv(tf.concat(inputs, 3))
class TFBottleneck(keras.layers.Layer):
# Standard bottleneck
def __init__(self, c1, c2, shortcut=True, g=1, e=0.5, w=None): # ch_in, ch_out, shortcut, groups, expansion
super().__init__()
c_ = int(c2 * e) # hidden channels
self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
self.cv2 = TFConv(c_, c2, 3, 1, g=g, w=w.cv2)
self.add = shortcut and c1 == c2
def call(self, inputs):
return inputs + self.cv2(self.cv1(inputs)) if self.add else self.cv2(self.cv1(inputs))
class TFCrossConv(keras.layers.Layer):
# Cross Convolution
def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False, w=None):
super().__init__()
c_ = int(c2 * e) # hidden channels
self.cv1 = TFConv(c1, c_, (1, k), (1, s), w=w.cv1)
self.cv2 = TFConv(c_, c2, (k, 1), (s, 1), g=g, w=w.cv2)
self.add = shortcut and c1 == c2
def call(self, inputs):
return inputs + self.cv2(self.cv1(inputs)) if self.add else self.cv2(self.cv1(inputs))
class TFConv2d(keras.layers.Layer):
# Substitution for PyTorch nn.Conv2D
def __init__(self, c1, c2, k, s=1, g=1, bias=True, w=None):
super().__init__()
assert g == 1, "TF v2.2 Conv2D does not support 'groups' argument"
self.conv = keras.layers.Conv2D(filters=c2,
kernel_size=k,
strides=s,
padding='VALID',
use_bias=bias,
kernel_initializer=keras.initializers.Constant(
w.weight.permute(2, 3, 1, 0).numpy()),
bias_initializer=keras.initializers.Constant(w.bias.numpy()) if bias else None)
def call(self, inputs):
return self.conv(inputs)
class TFBottleneckCSP(keras.layers.Layer):
# CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5, w=None):
# ch_in, ch_out, number, shortcut, groups, expansion
super().__init__()
c_ = int(c2 * e) # hidden channels
self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
self.cv2 = TFConv2d(c1, c_, 1, 1, bias=False, w=w.cv2)
self.cv3 = TFConv2d(c_, c_, 1, 1, bias=False, w=w.cv3)
self.cv4 = TFConv(2 * c_, c2, 1, 1, w=w.cv4)
self.bn = TFBN(w.bn)
self.act = lambda x: keras.activations.swish(x)
self.m = keras.Sequential([TFBottleneck(c_, c_, shortcut, g, e=1.0, w=w.m[j]) for j in range(n)])
def call(self, inputs):
y1 = self.cv3(self.m(self.cv1(inputs)))
y2 = self.cv2(inputs)
return self.cv4(self.act(self.bn(tf.concat((y1, y2), axis=3))))
class TFC3(keras.layers.Layer):
# CSP Bottleneck with 3 convolutions
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5, w=None):
# ch_in, ch_out, number, shortcut, groups, expansion
super().__init__()
c_ = int(c2 * e) # hidden channels
self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
self.cv2 = TFConv(c1, c_, 1, 1, w=w.cv2)
self.cv3 = TFConv(2 * c_, c2, 1, 1, w=w.cv3)
self.m = keras.Sequential([TFBottleneck(c_, c_, shortcut, g, e=1.0, w=w.m[j]) for j in range(n)])
def call(self, inputs):
return self.cv3(tf.concat((self.m(self.cv1(inputs)), self.cv2(inputs)), axis=3))
class TFC3x(keras.layers.Layer):
# 3 module with cross-convolutions
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5, w=None):
# ch_in, ch_out, number, shortcut, groups, expansion
super().__init__()
c_ = int(c2 * e) # hidden channels
self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
self.cv2 = TFConv(c1, c_, 1, 1, w=w.cv2)
self.cv3 = TFConv(2 * c_, c2, 1, 1, w=w.cv3)
self.m = keras.Sequential([
TFCrossConv(c_, c_, k=3, s=1, g=g, e=1.0, shortcut=shortcut, w=w.m[j]) for j in range(n)])
def call(self, inputs):
return self.cv3(tf.concat((self.m(self.cv1(inputs)), self.cv2(inputs)), axis=3))
class TFSPP(keras.layers.Layer):
# Spatial pyramid pooling layer used in YOLOv3-SPP
def __init__(self, c1, c2, k=(5, 9, 13), w=None):
super().__init__()
c_ = c1 // 2 # hidden channels
self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
self.cv2 = TFConv(c_ * (len(k) + 1), c2, 1, 1, w=w.cv2)
self.m = [keras.layers.MaxPool2D(pool_size=x, strides=1, padding='SAME') for x in k]
def call(self, inputs):
x = self.cv1(inputs)
return self.cv2(tf.concat([x] + [m(x) for m in self.m], 3))
class TFSPPF(keras.layers.Layer):
# Spatial pyramid pooling-Fast layer
def __init__(self, c1, c2, k=5, w=None):
super().__init__()
c_ = c1 // 2 # hidden channels
self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
self.cv2 = TFConv(c_ * 4, c2, 1, 1, w=w.cv2)
self.m = keras.layers.MaxPool2D(pool_size=k, strides=1, padding='SAME')
def call(self, inputs):
x = self.cv1(inputs)
y1 = self.m(x)
y2 = self.m(y1)
return self.cv2(tf.concat([x, y1, y2, self.m(y2)], 3))
class TFDetect(keras.layers.Layer):
# TF YOLO Detect layer
def __init__(self, nc=80, anchors=(), ch=(), imgsz=(640, 640), w=None): # detection layer
super().__init__()
self.stride = tf.convert_to_tensor(w.stride.numpy(), dtype=tf.float32)
self.nc = nc # number of classes
self.no = nc + 5 # number of outputs per anchor
self.nl = len(anchors) # number of detection layers
self.na = len(anchors[0]) // 2 # number of anchors
self.grid = [tf.zeros(1)] * self.nl # init grid
self.anchors = tf.convert_to_tensor(w.anchors.numpy(), dtype=tf.float32)
self.anchor_grid = tf.reshape(self.anchors * tf.reshape(self.stride, [self.nl, 1, 1]), [self.nl, 1, -1, 1, 2])
self.m = [TFConv2d(x, self.no * self.na, 1, w=w.m[i]) for i, x in enumerate(ch)]
self.training = False # set to False after building model
self.imgsz = imgsz
for i in range(self.nl):
ny, nx = self.imgsz[0] // self.stride[i], self.imgsz[1] // self.stride[i]
self.grid[i] = self._make_grid(nx, ny)
def call(self, inputs):
z = [] # inference output
x = []
for i in range(self.nl):
x.append(self.m[i](inputs[i]))
# x(bs,20,20,255) to x(bs,3,20,20,85)
ny, nx = self.imgsz[0] // self.stride[i], self.imgsz[1] // self.stride[i]
x[i] = tf.reshape(x[i], [-1, ny * nx, self.na, self.no])
if not self.training: # inference
y = x[i]
grid = tf.transpose(self.grid[i], [0, 2, 1, 3]) - 0.5
anchor_grid = tf.transpose(self.anchor_grid[i], [0, 2, 1, 3]) * 4
xy = (tf.sigmoid(y[..., 0:2]) * 2 + grid) * self.stride[i] # xy
wh = tf.sigmoid(y[..., 2:4]) ** 2 * anchor_grid
# Normalize xywh to 0-1 to reduce calibration error
xy /= tf.constant([[self.imgsz[1], self.imgsz[0]]], dtype=tf.float32)
wh /= tf.constant([[self.imgsz[1], self.imgsz[0]]], dtype=tf.float32)
y = tf.concat([xy, wh, tf.sigmoid(y[..., 4:5 + self.nc]), y[..., 5 + self.nc:]], -1)
z.append(tf.reshape(y, [-1, self.na * ny * nx, self.no]))
return tf.transpose(x, [0, 2, 1, 3]) if self.training else (tf.concat(z, 1),)
@staticmethod
def _make_grid(nx=20, ny=20):
# yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
# return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
xv, yv = tf.meshgrid(tf.range(nx), tf.range(ny))
return tf.cast(tf.reshape(tf.stack([xv, yv], 2), [1, 1, ny * nx, 2]), dtype=tf.float32)
class TFSegment(TFDetect):
# YOLO Segment head for segmentation models
def __init__(self, nc=80, anchors=(), nm=32, npr=256, ch=(), imgsz=(640, 640), w=None):
super().__init__(nc, anchors, ch, imgsz, w)
self.nm = nm # number of masks
self.npr = npr # number of protos
self.no = 5 + nc + self.nm # number of outputs per anchor
self.m = [TFConv2d(x, self.no * self.na, 1, w=w.m[i]) for i, x in enumerate(ch)] # output conv
self.proto = TFProto(ch[0], self.npr, self.nm, w=w.proto) # protos
self.detect = TFDetect.call
def call(self, x):
p = self.proto(x[0])
# p = TFUpsample(None, scale_factor=4, mode='nearest')(self.proto(x[0])) # (optional) full-size protos
p = tf.transpose(p, [0, 3, 1, 2]) # from shape(1,160,160,32) to shape(1,32,160,160)
x = self.detect(self, x)
return (x, p) if self.training else (x[0], p)
class TFProto(keras.layers.Layer):
def __init__(self, c1, c_=256, c2=32, w=None):
super().__init__()
self.cv1 = TFConv(c1, c_, k=3, w=w.cv1)
self.upsample = TFUpsample(None, scale_factor=2, mode='nearest')
self.cv2 = TFConv(c_, c_, k=3, w=w.cv2)
self.cv3 = TFConv(c_, c2, w=w.cv3)
def call(self, inputs):
return self.cv3(self.cv2(self.upsample(self.cv1(inputs))))
class TFUpsample(keras.layers.Layer):
# TF version of torch.nn.Upsample()
def __init__(self, size, scale_factor, mode, w=None): # warning: all arguments needed including 'w'
super().__init__()
assert scale_factor % 2 == 0, "scale_factor must be multiple of 2"
self.upsample = lambda x: tf.image.resize(x, (x.shape[1] * scale_factor, x.shape[2] * scale_factor), mode)
# self.upsample = keras.layers.UpSampling2D(size=scale_factor, interpolation=mode)
# with default arguments: align_corners=False, half_pixel_centers=False
# self.upsample = lambda x: tf.raw_ops.ResizeNearestNeighbor(images=x,
# size=(x.shape[1] * 2, x.shape[2] * 2))
def call(self, inputs):
return self.upsample(inputs)
class TFConcat(keras.layers.Layer):
# TF version of torch.concat()
def __init__(self, dimension=1, w=None):
super().__init__()
assert dimension == 1, "convert only NCHW to NHWC concat"
self.d = 3
def call(self, inputs):
return tf.concat(inputs, self.d)
def parse_model(d, ch, model, imgsz): # model_dict, input_channels(3)
LOGGER.info(f"\n{'':>3}{'from':>18}{'n':>3}{'params':>10} {'module':<40}{'arguments':<30}")
anchors, nc, gd, gw = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple']
na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors # number of anchors
no = na * (nc + 5) # number of outputs = anchors * (classes + 5)
layers, save, c2 = [], [], ch[-1] # layers, savelist, ch out
for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']): # from, number, module, args
m_str = m
m = eval(m) if isinstance(m, str) else m # eval strings
for j, a in enumerate(args):
try:
args[j] = eval(a) if isinstance(a, str) else a # eval strings
except NameError:
pass
n = max(round(n * gd), 1) if n > 1 else n # depth gain
if m in [
nn.Conv2d, Conv, DWConv, DWConvTranspose2d, Bottleneck, SPP, SPPF, MixConv2d, Focus, CrossConv,
BottleneckCSP, C3, C3x]:
c1, c2 = ch[f], args[0]
c2 = make_divisible(c2 * gw, 8) if c2 != no else c2
args = [c1, c2, *args[1:]]
if m in [BottleneckCSP, C3, C3x]:
args.insert(2, n)
n = 1
elif m is nn.BatchNorm2d:
args = [ch[f]]
elif m is Concat:
c2 = sum(ch[-1 if x == -1 else x + 1] for x in f)
elif m in [Detect, Segment]:
args.append([ch[x + 1] for x in f])
if isinstance(args[1], int): # number of anchors
args[1] = [list(range(args[1] * 2))] * len(f)
if m is Segment:
args[3] = make_divisible(args[3] * gw, 8)
args.append(imgsz)
else:
c2 = ch[f]
tf_m = eval('TF' + m_str.replace('nn.', ''))
m_ = keras.Sequential([tf_m(*args, w=model.model[i][j]) for j in range(n)]) if n > 1 \
else tf_m(*args, w=model.model[i]) # module
torch_m_ = nn.Sequential(*(m(*args) for _ in range(n))) if n > 1 else m(*args) # module
t = str(m)[8:-2].replace('__main__.', '') # module type
np = sum(x.numel() for x in torch_m_.parameters()) # number params
m_.i, m_.f, m_.type, m_.np = i, f, t, np # attach index, 'from' index, type, number params
LOGGER.info(f'{i:>3}{str(f):>18}{str(n):>3}{np:>10} {t:<40}{str(args):<30}') # print
save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1) # append to savelist
layers.append(m_)
ch.append(c2)
return keras.Sequential(layers), sorted(save)
class TFModel:
# TF YOLO model
def __init__(self, cfg='yolo.yaml', ch=3, nc=None, model=None, imgsz=(640, 640)): # model, channels, classes
super().__init__()
if isinstance(cfg, dict):
self.yaml = cfg # model dict
else: # is *.yaml
import yaml # for torch hub
self.yaml_file = Path(cfg).name
with open(cfg) as f:
self.yaml = yaml.load(f, Loader=yaml.FullLoader) # model dict
# Define model
if nc and nc != self.yaml['nc']:
LOGGER.info(f"Overriding {cfg} nc={self.yaml['nc']} with nc={nc}")
self.yaml['nc'] = nc # override yaml value
self.model, self.savelist = parse_model(deepcopy(self.yaml), ch=[ch], model=model, imgsz=imgsz)
def predict(self,
inputs,
tf_nms=False,
agnostic_nms=False,
topk_per_class=100,
topk_all=100,
iou_thres=0.45,
conf_thres=0.25):
y = [] # outputs
x = inputs
for m in self.model.layers:
if m.f != -1: # if not from previous layer
x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f] # from earlier layers
x = m(x) # run
y.append(x if m.i in self.savelist else None) # save output
# Add TensorFlow NMS
if tf_nms:
boxes = self._xywh2xyxy(x[0][..., :4])
probs = x[0][:, :, 4:5]
classes = x[0][:, :, 5:]
scores = probs * classes
if agnostic_nms:
nms = AgnosticNMS()((boxes, classes, scores), topk_all, iou_thres, conf_thres)
else:
boxes = tf.expand_dims(boxes, 2)
nms = tf.image.combined_non_max_suppression(boxes,
scores,
topk_per_class,
topk_all,
iou_thres,
conf_thres,
clip_boxes=False)
return (nms,)
return x # output [1,6300,85] = [xywh, conf, class0, class1, ...]
# x = x[0] # [x(1,6300,85), ...] to x(6300,85)
# xywh = x[..., :4] # x(6300,4) boxes
# conf = x[..., 4:5] # x(6300,1) confidences
# cls = tf.reshape(tf.cast(tf.argmax(x[..., 5:], axis=1), tf.float32), (-1, 1)) # x(6300,1) classes
# return tf.concat([conf, cls, xywh], 1)
@staticmethod
def _xywh2xyxy(xywh):
# Convert nx4 boxes from [x, y, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
x, y, w, h = tf.split(xywh, num_or_size_splits=4, axis=-1)
return tf.concat([x - w / 2, y - h / 2, x + w / 2, y + h / 2], axis=-1)
class AgnosticNMS(keras.layers.Layer):
# TF Agnostic NMS
def call(self, input, topk_all, iou_thres, conf_thres):
# wrap map_fn to avoid TypeSpec related error https://stackoverflow.com/a/65809989/3036450
return tf.map_fn(lambda x: self._nms(x, topk_all, iou_thres, conf_thres),
input,
fn_output_signature=(tf.float32, tf.float32, tf.float32, tf.int32),
name='agnostic_nms')
@staticmethod
def _nms(x, topk_all=100, iou_thres=0.45, conf_thres=0.25): # agnostic NMS
boxes, classes, scores = x
class_inds = tf.cast(tf.argmax(classes, axis=-1), tf.float32)
scores_inp = tf.reduce_max(scores, -1)
selected_inds = tf.image.non_max_suppression(boxes,
scores_inp,
max_output_size=topk_all,
iou_threshold=iou_thres,
score_threshold=conf_thres)
selected_boxes = tf.gather(boxes, selected_inds)
padded_boxes = tf.pad(selected_boxes,
paddings=[[0, topk_all - tf.shape(selected_boxes)[0]], [0, 0]],
mode="CONSTANT",
constant_values=0.0)
selected_scores = tf.gather(scores_inp, selected_inds)
padded_scores = tf.pad(selected_scores,
paddings=[[0, topk_all - tf.shape(selected_boxes)[0]]],
mode="CONSTANT",
constant_values=-1.0)
selected_classes = tf.gather(class_inds, selected_inds)
padded_classes = tf.pad(selected_classes,
paddings=[[0, topk_all - tf.shape(selected_boxes)[0]]],
mode="CONSTANT",
constant_values=-1.0)
valid_detections = tf.shape(selected_inds)[0]
return padded_boxes, padded_scores, padded_classes, valid_detections
def activations(act=nn.SiLU):
# Returns TF activation from input PyTorch activation
if isinstance(act, nn.LeakyReLU):
return lambda x: keras.activations.relu(x, alpha=0.1)
elif isinstance(act, nn.Hardswish):
return lambda x: x * tf.nn.relu6(x + 3) * 0.166666667
elif isinstance(act, (nn.SiLU, SiLU)):
return lambda x: keras.activations.swish(x)
else:
raise Exception(f'no matching TensorFlow activation found for PyTorch activation {act}')
def representative_dataset_gen(dataset, ncalib=100):
# Representative dataset generator for use with converter.representative_dataset, returns a generator of np arrays
for n, (path, img, im0s, vid_cap, string) in enumerate(dataset):
im = np.transpose(img, [1, 2, 0])
im = np.expand_dims(im, axis=0).astype(np.float32)
im /= 255
yield [im]
if n >= ncalib:
break
def run(
weights=ROOT / 'yolo.pt', # weights path
imgsz=(640, 640), # inference size h,w
batch_size=1, # batch size
dynamic=False, # dynamic batch size
):
# PyTorch model
im = torch.zeros((batch_size, 3, *imgsz)) # BCHW image
model = attempt_load(weights, device=torch.device('cpu'), inplace=True, fuse=False)
_ = model(im) # inference
model.info()
# TensorFlow model
im = tf.zeros((batch_size, *imgsz, 3)) # BHWC image
tf_model = TFModel(cfg=model.yaml, model=model, nc=model.nc, imgsz=imgsz)
_ = tf_model.predict(im) # inference
# Keras model
im = keras.Input(shape=(*imgsz, 3), batch_size=None if dynamic else batch_size)
keras_model = keras.Model(inputs=im, outputs=tf_model.predict(im))
keras_model.summary()
LOGGER.info('PyTorch, TensorFlow and Keras models successfully verified.\nUse export.py for TF model export.')
def parse_opt():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', type=str, default=ROOT / 'yolo.pt', help='weights path')
parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[640], help='inference size h,w')
parser.add_argument('--batch-size', type=int, default=1, help='batch size')
parser.add_argument('--dynamic', action='store_true', help='dynamic batch size')
opt = parser.parse_args()
opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1 # expand
print_args(vars(opt))
return opt
def main(opt):
run(**vars(opt))
if __name__ == "__main__":
opt = parse_opt()
main(opt)
models/yolo.py 0 → 100644
View file @ f969ca34
import argparse
import os
import platform
import sys
from copy import deepcopy
from pathlib import Path
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
if platform.system() != 'Windows':
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
from models.common import *
from models.experimental import *
from utils.general import LOGGER, check_version, check_yaml, make_divisible, print_args
from utils.plots import feature_visualization
from utils.torch_utils import (fuse_conv_and_bn, initialize_weights, model_info, profile, scale_img, select_device,
time_sync)
from utils.tal.anchor_generator import make_anchors, dist2bbox
try:
import thop # for FLOPs computation
except ImportError:
thop = None
class Detect(nn.Module):
# YOLO Detect head for detection models
dynamic = False # force grid reconstruction
export = False # export mode
shape = None
anchors = torch.empty(0) # init
strides = torch.empty(0) # init
def __init__(self, nc=80, ch=(), inplace=True): # detection layer
super().__init__()
self.nc = nc # number of classes
self.nl = len(ch) # number of detection layers
self.reg_max = 16
self.no = nc + self.reg_max * 4 # number of outputs per anchor
self.inplace = inplace # use inplace ops (e.g. slice assignment)
self.stride = torch.zeros(self.nl) # strides computed during build
c2, c3 = max((ch[0] // 4, self.reg_max * 4, 16)), max((ch[0], min((self.nc * 2, 128)))) # channels
self.cv2 = nn.ModuleList(
nn.Sequential(Conv(x, c2, 3), Conv(c2, c2, 3), nn.Conv2d(c2, 4 * self.reg_max, 1)) for x in ch)
self.cv3 = nn.ModuleList(
nn.Sequential(Conv(x, c3, 3), Conv(c3, c3, 3), nn.Conv2d(c3, self.nc, 1)) for x in ch)
self.dfl = DFL(self.reg_max) if self.reg_max > 1 else nn.Identity()
def forward(self, x):
shape = x[0].shape # BCHW
for i in range(self.nl):
x[i] = torch.cat((self.cv2[i](x[i]), self.cv3[i](x[i])), 1)
if self.training:
return x
elif self.dynamic or self.shape != shape:
self.anchors, self.strides = (x.transpose(0, 1) for x in make_anchors(x, self.stride, 0.5))
self.shape = shape
box, cls = torch.cat([xi.view(shape[0], self.no, -1) for xi in x], 2).split((self.reg_max * 4, self.nc), 1)
dbox = dist2bbox(self.dfl(box), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
y = torch.cat((dbox, cls.sigmoid()), 1)
return y if self.export else (y, x)
def bias_init(self):
# Initialize Detect() biases, WARNING: requires stride availability
m = self # self.model[-1] # Detect() module
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1
# ncf = math.log(0.6 / (m.nc - 0.999999)) if cf is None else torch.log(cf / cf.sum()) # nominal class frequency
for a, b, s in zip(m.cv2, m.cv3, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
class DDetect(nn.Module):
# YOLO Detect head for detection models
dynamic = False # force grid reconstruction
export = False # export mode
shape = None
anchors = torch.empty(0) # init
strides = torch.empty(0) # init
def __init__(self, nc=80, ch=(), inplace=True): # detection layer
super().__init__()
self.nc = nc # number of classes
self.nl = len(ch) # number of detection layers
self.reg_max = 16
self.no = nc + self.reg_max * 4 # number of outputs per anchor
self.inplace = inplace # use inplace ops (e.g. slice assignment)
self.stride = torch.zeros(self.nl) # strides computed during build
c2, c3 = make_divisible(max((ch[0] // 4, self.reg_max * 4, 16)), 4), max((ch[0], min((self.nc * 2, 128)))) # channels
self.cv2 = nn.ModuleList(
nn.Sequential(Conv(x, c2, 3), Conv(c2, c2, 3, g=4), nn.Conv2d(c2, 4 * self.reg_max, 1, groups=4)) for x in ch)
self.cv3 = nn.ModuleList(
nn.Sequential(Conv(x, c3, 3), Conv(c3, c3, 3), nn.Conv2d(c3, self.nc, 1)) for x in ch)
self.dfl = DFL(self.reg_max) if self.reg_max > 1 else nn.Identity()
def forward(self, x):
shape = x[0].shape # BCHW
for i in range(self.nl):
x[i] = torch.cat((self.cv2[i](x[i]), self.cv3[i](x[i])), 1)
if self.training:
return x
elif self.dynamic or self.shape != shape:
self.anchors, self.strides = (x.transpose(0, 1) for x in make_anchors(x, self.stride, 0.5))
self.shape = shape
box, cls = torch.cat([xi.view(shape[0], self.no, -1) for xi in x], 2).split((self.reg_max * 4, self.nc), 1)
dbox = dist2bbox(self.dfl(box), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
y = torch.cat((dbox, cls.sigmoid()), 1)
return y if self.export else (y, x)
def bias_init(self):
# Initialize Detect() biases, WARNING: requires stride availability
m = self # self.model[-1] # Detect() module
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1
# ncf = math.log(0.6 / (m.nc - 0.999999)) if cf is None else torch.log(cf / cf.sum()) # nominal class frequency
for a, b, s in zip(m.cv2, m.cv3, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
class DualDetect(nn.Module):
# YOLO Detect head for detection models
dynamic = False # force grid reconstruction
export = False # export mode
shape = None
anchors = torch.empty(0) # init
strides = torch.empty(0) # init
def __init__(self, nc=80, ch=(), inplace=True): # detection layer
super().__init__()
self.nc = nc # number of classes
self.nl = len(ch) // 2 # number of detection layers
self.reg_max = 16
self.no = nc + self.reg_max * 4 # number of outputs per anchor
self.inplace = inplace # use inplace ops (e.g. slice assignment)
self.stride = torch.zeros(self.nl) # strides computed during build
c2, c3 = max((ch[0] // 4, self.reg_max * 4, 16)), max((ch[0], min((self.nc * 2, 128)))) # channels
c4, c5 = max((ch[self.nl] // 4, self.reg_max * 4, 16)), max((ch[self.nl], min((self.nc * 2, 128)))) # channels
self.cv2 = nn.ModuleList(
nn.Sequential(Conv(x, c2, 3), Conv(c2, c2, 3), nn.Conv2d(c2, 4 * self.reg_max, 1)) for x in ch[:self.nl])
self.cv3 = nn.ModuleList(
nn.Sequential(Conv(x, c3, 3), Conv(c3, c3, 3), nn.Conv2d(c3, self.nc, 1)) for x in ch[:self.nl])
self.cv4 = nn.ModuleList(
nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3), nn.Conv2d(c4, 4 * self.reg_max, 1)) for x in ch[self.nl:])
self.cv5 = nn.ModuleList(
nn.Sequential(Conv(x, c5, 3), Conv(c5, c5, 3), nn.Conv2d(c5, self.nc, 1)) for x in ch[self.nl:])
self.dfl = DFL(self.reg_max)
self.dfl2 = DFL(self.reg_max)
def forward(self, x):
shape = x[0].shape # BCHW
d1 = []
d2 = []
for i in range(self.nl):
d1.append(torch.cat((self.cv2[i](x[i]), self.cv3[i](x[i])), 1))
d2.append(torch.cat((self.cv4[i](x[self.nl+i]), self.cv5[i](x[self.nl+i])), 1))
if self.training:
return [d1, d2]
elif self.dynamic or self.shape != shape:
self.anchors, self.strides = (d1.transpose(0, 1) for d1 in make_anchors(d1, self.stride, 0.5))
self.shape = shape
box, cls = torch.cat([di.view(shape[0], self.no, -1) for di in d1], 2).split((self.reg_max * 4, self.nc), 1)
dbox = dist2bbox(self.dfl(box), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
box2, cls2 = torch.cat([di.view(shape[0], self.no, -1) for di in d2], 2).split((self.reg_max * 4, self.nc), 1)
dbox2 = dist2bbox(self.dfl2(box2), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
y = [torch.cat((dbox, cls.sigmoid()), 1), torch.cat((dbox2, cls2.sigmoid()), 1)]
return y if self.export else (y, [d1, d2])
def bias_init(self):
# Initialize Detect() biases, WARNING: requires stride availability
m = self # self.model[-1] # Detect() module
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1
# ncf = math.log(0.6 / (m.nc - 0.999999)) if cf is None else torch.log(cf / cf.sum()) # nominal class frequency
for a, b, s in zip(m.cv2, m.cv3, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
for a, b, s in zip(m.cv4, m.cv5, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
class DualDDetect(nn.Module):
# YOLO Detect head for detection models
dynamic = False # force grid reconstruction
export = False # export mode
shape = None
anchors = torch.empty(0) # init
strides = torch.empty(0) # init
def __init__(self, nc=80, ch=(), inplace=True): # detection layer
super().__init__()
self.nc = nc # number of classes
self.nl = len(ch) // 2 # number of detection layers
self.reg_max = 16
self.no = nc + self.reg_max * 4 # number of outputs per anchor
self.inplace = inplace # use inplace ops (e.g. slice assignment)
self.stride = torch.zeros(self.nl) # strides computed during build
c2, c3 = make_divisible(max((ch[0] // 4, self.reg_max * 4, 16)), 4), max((ch[0], min((self.nc * 2, 128)))) # channels
c4, c5 = make_divisible(max((ch[self.nl] // 4, self.reg_max * 4, 16)), 4), max((ch[self.nl], min((self.nc * 2, 128)))) # channels
self.cv2 = nn.ModuleList(
nn.Sequential(Conv(x, c2, 3), Conv(c2, c2, 3, g=4), nn.Conv2d(c2, 4 * self.reg_max, 1, groups=4)) for x in ch[:self.nl])
self.cv3 = nn.ModuleList(
nn.Sequential(Conv(x, c3, 3), Conv(c3, c3, 3), nn.Conv2d(c3, self.nc, 1)) for x in ch[:self.nl])
self.cv4 = nn.ModuleList(
nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3, g=4), nn.Conv2d(c4, 4 * self.reg_max, 1, groups=4)) for x in ch[self.nl:])
self.cv5 = nn.ModuleList(
nn.Sequential(Conv(x, c5, 3), Conv(c5, c5, 3), nn.Conv2d(c5, self.nc, 1)) for x in ch[self.nl:])
self.dfl = DFL(self.reg_max)
self.dfl2 = DFL(self.reg_max)
def forward(self, x):
shape = x[0].shape # BCHW
d1 = []
d2 = []
for i in range(self.nl):
d1.append(torch.cat((self.cv2[i](x[i]), self.cv3[i](x[i])), 1))
d2.append(torch.cat((self.cv4[i](x[self.nl+i]), self.cv5[i](x[self.nl+i])), 1))
if self.training:
return [d1, d2]
elif self.dynamic or self.shape != shape:
self.anchors, self.strides = (d1.transpose(0, 1) for d1 in make_anchors(d1, self.stride, 0.5))
self.shape = shape
box, cls = torch.cat([di.view(shape[0], self.no, -1) for di in d1], 2).split((self.reg_max * 4, self.nc), 1)
dbox = dist2bbox(self.dfl(box), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
box2, cls2 = torch.cat([di.view(shape[0], self.no, -1) for di in d2], 2).split((self.reg_max * 4, self.nc), 1)
dbox2 = dist2bbox(self.dfl2(box2), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
y = [torch.cat((dbox, cls.sigmoid()), 1), torch.cat((dbox2, cls2.sigmoid()), 1)]
return y if self.export else (y, [d1, d2])
#y = torch.cat((dbox2, cls2.sigmoid()), 1)
#return y if self.export else (y, d2)
#y1 = torch.cat((dbox, cls.sigmoid()), 1)
#y2 = torch.cat((dbox2, cls2.sigmoid()), 1)
#return [y1, y2] if self.export else [(y1, d1), (y2, d2)]
#return [y1, y2] if self.export else [(y1, y2), (d1, d2)]
def bias_init(self):
# Initialize Detect() biases, WARNING: requires stride availability
m = self # self.model[-1] # Detect() module
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1
# ncf = math.log(0.6 / (m.nc - 0.999999)) if cf is None else torch.log(cf / cf.sum()) # nominal class frequency
for a, b, s in zip(m.cv2, m.cv3, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
for a, b, s in zip(m.cv4, m.cv5, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
class TripleDetect(nn.Module):
# YOLO Detect head for detection models
dynamic = False # force grid reconstruction
export = False # export mode
shape = None
anchors = torch.empty(0) # init
strides = torch.empty(0) # init
def __init__(self, nc=80, ch=(), inplace=True): # detection layer
super().__init__()
self.nc = nc # number of classes
self.nl = len(ch) // 3 # number of detection layers
self.reg_max = 16
self.no = nc + self.reg_max * 4 # number of outputs per anchor
self.inplace = inplace # use inplace ops (e.g. slice assignment)
self.stride = torch.zeros(self.nl) # strides computed during build
c2, c3 = max((ch[0] // 4, self.reg_max * 4, 16)), max((ch[0], min((self.nc * 2, 128)))) # channels
c4, c5 = max((ch[self.nl] // 4, self.reg_max * 4, 16)), max((ch[self.nl], min((self.nc * 2, 128)))) # channels
c6, c7 = max((ch[self.nl * 2] // 4, self.reg_max * 4, 16)), max((ch[self.nl * 2], min((self.nc * 2, 128)))) # channels
self.cv2 = nn.ModuleList(
nn.Sequential(Conv(x, c2, 3), Conv(c2, c2, 3), nn.Conv2d(c2, 4 * self.reg_max, 1)) for x in ch[:self.nl])
self.cv3 = nn.ModuleList(
nn.Sequential(Conv(x, c3, 3), Conv(c3, c3, 3), nn.Conv2d(c3, self.nc, 1)) for x in ch[:self.nl])
self.cv4 = nn.ModuleList(
nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3), nn.Conv2d(c4, 4 * self.reg_max, 1)) for x in ch[self.nl:self.nl*2])
self.cv5 = nn.ModuleList(
nn.Sequential(Conv(x, c5, 3), Conv(c5, c5, 3), nn.Conv2d(c5, self.nc, 1)) for x in ch[self.nl:self.nl*2])
self.cv6 = nn.ModuleList(
nn.Sequential(Conv(x, c6, 3), Conv(c6, c6, 3), nn.Conv2d(c6, 4 * self.reg_max, 1)) for x in ch[self.nl*2:self.nl*3])
self.cv7 = nn.ModuleList(
nn.Sequential(Conv(x, c7, 3), Conv(c7, c7, 3), nn.Conv2d(c7, self.nc, 1)) for x in ch[self.nl*2:self.nl*3])
self.dfl = DFL(self.reg_max)
self.dfl2 = DFL(self.reg_max)
self.dfl3 = DFL(self.reg_max)
def forward(self, x):
shape = x[0].shape # BCHW
d1 = []
d2 = []
d3 = []
for i in range(self.nl):
d1.append(torch.cat((self.cv2[i](x[i]), self.cv3[i](x[i])), 1))
d2.append(torch.cat((self.cv4[i](x[self.nl+i]), self.cv5[i](x[self.nl+i])), 1))
d3.append(torch.cat((self.cv6[i](x[self.nl*2+i]), self.cv7[i](x[self.nl*2+i])), 1))
if self.training:
return [d1, d2, d3]
elif self.dynamic or self.shape != shape:
self.anchors, self.strides = (d1.transpose(0, 1) for d1 in make_anchors(d1, self.stride, 0.5))
self.shape = shape
box, cls = torch.cat([di.view(shape[0], self.no, -1) for di in d1], 2).split((self.reg_max * 4, self.nc), 1)
dbox = dist2bbox(self.dfl(box), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
box2, cls2 = torch.cat([di.view(shape[0], self.no, -1) for di in d2], 2).split((self.reg_max * 4, self.nc), 1)
dbox2 = dist2bbox(self.dfl2(box2), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
box3, cls3 = torch.cat([di.view(shape[0], self.no, -1) for di in d3], 2).split((self.reg_max * 4, self.nc), 1)
dbox3 = dist2bbox(self.dfl3(box3), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
y = [torch.cat((dbox, cls.sigmoid()), 1), torch.cat((dbox2, cls2.sigmoid()), 1), torch.cat((dbox3, cls3.sigmoid()), 1)]
return y if self.export else (y, [d1, d2, d3])
def bias_init(self):
# Initialize Detect() biases, WARNING: requires stride availability
m = self # self.model[-1] # Detect() module
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1
# ncf = math.log(0.6 / (m.nc - 0.999999)) if cf is None else torch.log(cf / cf.sum()) # nominal class frequency
for a, b, s in zip(m.cv2, m.cv3, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
for a, b, s in zip(m.cv4, m.cv5, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
for a, b, s in zip(m.cv6, m.cv7, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
class TripleDDetect(nn.Module):
# YOLO Detect head for detection models
dynamic = False # force grid reconstruction
export = False # export mode
shape = None
anchors = torch.empty(0) # init
strides = torch.empty(0) # init
def __init__(self, nc=80, ch=(), inplace=True): # detection layer
super().__init__()
self.nc = nc # number of classes
self.nl = len(ch) // 3 # number of detection layers
self.reg_max = 16
self.no = nc + self.reg_max * 4 # number of outputs per anchor
self.inplace = inplace # use inplace ops (e.g. slice assignment)
self.stride = torch.zeros(self.nl) # strides computed during build
c2, c3 = make_divisible(max((ch[0] // 4, self.reg_max * 4, 16)), 4), \
max((ch[0], min((self.nc * 2, 128)))) # channels
c4, c5 = make_divisible(max((ch[self.nl] // 4, self.reg_max * 4, 16)), 4), \
max((ch[self.nl], min((self.nc * 2, 128)))) # channels
c6, c7 = make_divisible(max((ch[self.nl * 2] // 4, self.reg_max * 4, 16)), 4), \
max((ch[self.nl * 2], min((self.nc * 2, 128)))) # channels
self.cv2 = nn.ModuleList(
nn.Sequential(Conv(x, c2, 3), Conv(c2, c2, 3, g=4),
nn.Conv2d(c2, 4 * self.reg_max, 1, groups=4)) for x in ch[:self.nl])
self.cv3 = nn.ModuleList(
nn.Sequential(Conv(x, c3, 3), Conv(c3, c3, 3), nn.Conv2d(c3, self.nc, 1)) for x in ch[:self.nl])
self.cv4 = nn.ModuleList(
nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3, g=4),
nn.Conv2d(c4, 4 * self.reg_max, 1, groups=4)) for x in ch[self.nl:self.nl*2])
self.cv5 = nn.ModuleList(
nn.Sequential(Conv(x, c5, 3), Conv(c5, c5, 3), nn.Conv2d(c5, self.nc, 1)) for x in ch[self.nl:self.nl*2])
self.cv6 = nn.ModuleList(
nn.Sequential(Conv(x, c6, 3), Conv(c6, c6, 3, g=4),
nn.Conv2d(c6, 4 * self.reg_max, 1, groups=4)) for x in ch[self.nl*2:self.nl*3])
self.cv7 = nn.ModuleList(
nn.Sequential(Conv(x, c7, 3), Conv(c7, c7, 3), nn.Conv2d(c7, self.nc, 1)) for x in ch[self.nl*2:self.nl*3])
self.dfl = DFL(self.reg_max)
self.dfl2 = DFL(self.reg_max)
self.dfl3 = DFL(self.reg_max)
def forward(self, x):
shape = x[0].shape # BCHW
d1 = []
d2 = []
d3 = []
for i in range(self.nl):
d1.append(torch.cat((self.cv2[i](x[i]), self.cv3[i](x[i])), 1))
d2.append(torch.cat((self.cv4[i](x[self.nl+i]), self.cv5[i](x[self.nl+i])), 1))
d3.append(torch.cat((self.cv6[i](x[self.nl*2+i]), self.cv7[i](x[self.nl*2+i])), 1))
if self.training:
return [d1, d2, d3]
elif self.dynamic or self.shape != shape:
self.anchors, self.strides = (d1.transpose(0, 1) for d1 in make_anchors(d1, self.stride, 0.5))
self.shape = shape
box, cls = torch.cat([di.view(shape[0], self.no, -1) for di in d1], 2).split((self.reg_max * 4, self.nc), 1)
dbox = dist2bbox(self.dfl(box), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
box2, cls2 = torch.cat([di.view(shape[0], self.no, -1) for di in d2], 2).split((self.reg_max * 4, self.nc), 1)
dbox2 = dist2bbox(self.dfl2(box2), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
box3, cls3 = torch.cat([di.view(shape[0], self.no, -1) for di in d3], 2).split((self.reg_max * 4, self.nc), 1)
dbox3 = dist2bbox(self.dfl3(box3), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
#y = [torch.cat((dbox, cls.sigmoid()), 1), torch.cat((dbox2, cls2.sigmoid()), 1), torch.cat((dbox3, cls3.sigmoid()), 1)]
#return y if self.export else (y, [d1, d2, d3])
y = torch.cat((dbox3, cls3.sigmoid()), 1)
return y if self.export else (y, d3)
def bias_init(self):
# Initialize Detect() biases, WARNING: requires stride availability
m = self # self.model[-1] # Detect() module
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1
# ncf = math.log(0.6 / (m.nc - 0.999999)) if cf is None else torch.log(cf / cf.sum()) # nominal class frequency
for a, b, s in zip(m.cv2, m.cv3, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
for a, b, s in zip(m.cv4, m.cv5, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
for a, b, s in zip(m.cv6, m.cv7, m.stride): # from
a[-1].bias.data[:] = 1.0 # box
b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2) # cls (5 objects and 80 classes per 640 image)
class Segment(Detect):
# YOLO Segment head for segmentation models
def __init__(self, nc=80, nm=32, npr=256, ch=(), inplace=True):
super().__init__(nc, ch, inplace)
self.nm = nm # number of masks
self.npr = npr # number of protos
self.proto = Proto(ch[0], self.npr, self.nm) # protos
self.detect = Detect.forward
c4 = max(ch[0] // 4, self.nm)
self.cv4 = nn.ModuleList(nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3), nn.Conv2d(c4, self.nm, 1)) for x in ch)
def forward(self, x):
p = self.proto(x[0])
bs = p.shape[0]
mc = torch.cat([self.cv4[i](x[i]).view(bs, self.nm, -1) for i in range(self.nl)], 2) # mask coefficients
x = self.detect(self, x)
if self.training:
return x, mc, p
return (torch.cat([x, mc], 1), p) if self.export else (torch.cat([x[0], mc], 1), (x[1], mc, p))
class DSegment(DDetect):
# YOLO Segment head for segmentation models
def __init__(self, nc=80, nm=32, npr=256, ch=(), inplace=True):
super().__init__(nc, ch[:-1], inplace)
self.nl = len(ch)-1
self.nm = nm # number of masks
self.npr = npr # number of protos
self.proto = Conv(ch[-1], self.nm, 1) # protos
self.detect = DDetect.forward
c4 = max(ch[0] // 4, self.nm)
self.cv4 = nn.ModuleList(nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3), nn.Conv2d(c4, self.nm, 1)) for x in ch[:-1])
def forward(self, x):
p = self.proto(x[-1])
bs = p.shape[0]
mc = torch.cat([self.cv4[i](x[i]).view(bs, self.nm, -1) for i in range(self.nl)], 2) # mask coefficients
x = self.detect(self, x[:-1])
if self.training:
return x, mc, p
return (torch.cat([x, mc], 1), p) if self.export else (torch.cat([x[0], mc], 1), (x[1], mc, p))
class DualDSegment(DualDDetect):
# YOLO Segment head for segmentation models
def __init__(self, nc=80, nm=32, npr=256, ch=(), inplace=True):
super().__init__(nc, ch[:-2], inplace)
self.nl = (len(ch)-2) // 2
self.nm = nm # number of masks
self.npr = npr # number of protos
self.proto = Conv(ch[-2], self.nm, 1) # protos
self.proto2 = Conv(ch[-1], self.nm, 1) # protos
self.detect = DualDDetect.forward
c6 = max(ch[0] // 4, self.nm)
c7 = max(ch[self.nl] // 4, self.nm)
self.cv6 = nn.ModuleList(nn.Sequential(Conv(x, c6, 3), Conv(c6, c6, 3), nn.Conv2d(c6, self.nm, 1)) for x in ch[:self.nl])
self.cv7 = nn.ModuleList(nn.Sequential(Conv(x, c7, 3), Conv(c7, c7, 3), nn.Conv2d(c7, self.nm, 1)) for x in ch[self.nl:self.nl*2])
def forward(self, x):
p = [self.proto(x[-2]), self.proto2(x[-1])]
bs = p[0].shape[0]
mc = [torch.cat([self.cv6[i](x[i]).view(bs, self.nm, -1) for i in range(self.nl)], 2),
torch.cat([self.cv7[i](x[self.nl+i]).view(bs, self.nm, -1) for i in range(self.nl)], 2)] # mask coefficients
d = self.detect(self, x[:-2])
if self.training:
return d, mc, p
return (torch.cat([d[0][1], mc[1]], 1), (d[1][1], mc[1], p[1]))
class Panoptic(Detect):
# YOLO Panoptic head for panoptic segmentation models
def __init__(self, nc=80, sem_nc=93, nm=32, npr=256, ch=(), inplace=True):
super().__init__(nc, ch, inplace)
self.sem_nc = sem_nc
self.nm = nm # number of masks
self.npr = npr # number of protos
self.proto = Proto(ch[0], self.npr, self.nm) # protos
self.uconv = UConv(ch[0], ch[0]//4, self.sem_nc+self.nc)
self.detect = Detect.forward
c4 = max(ch[0] // 4, self.nm)
self.cv4 = nn.ModuleList(nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3), nn.Conv2d(c4, self.nm, 1)) for x in ch)
def forward(self, x):
p = self.proto(x[0])
s = self.uconv(x[0])
bs = p.shape[0]
mc = torch.cat([self.cv4[i](x[i]).view(bs, self.nm, -1) for i in range(self.nl)], 2) # mask coefficients
x = self.detect(self, x)
if self.training:
return x, mc, p, s
return (torch.cat([x, mc], 1), p, s) if self.export else (torch.cat([x[0], mc], 1), (x[1], mc, p, s))
class BaseModel(nn.Module):
# YOLO base model
def forward(self, x, profile=False, visualize=False):
return self._forward_once(x, profile, visualize) # single-scale inference, train
def _forward_once(self, x, profile=False, visualize=False):
y, dt = [], [] # outputs
for m in self.model:
if m.f != -1: # if not from previous layer
x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f] # from earlier layers
if profile:
self._profile_one_layer(m, x, dt)
x = m(x) # run
y.append(x if m.i in self.save else None) # save output
if visualize:
feature_visualization(x, m.type, m.i, save_dir=visualize)
return x
def _profile_one_layer(self, m, x, dt):
c = m == self.model[-1] # is final layer, copy input as inplace fix
o = thop.profile(m, inputs=(x.copy() if c else x,), verbose=False)[0] / 1E9 * 2 if thop else 0 # FLOPs
t = time_sync()
for _ in range(10):
m(x.copy() if c else x)
dt.append((time_sync() - t) * 100)
if m == self.model[0]:
LOGGER.info(f"{'time (ms)':>10s} {'GFLOPs':>10s} {'params':>10s} module")
LOGGER.info(f'{dt[-1]:10.2f} {o:10.2f} {m.np:10.0f} {m.type}')
if c:
LOGGER.info(f"{sum(dt):10.2f} {'-':>10s} {'-':>10s} Total")
def fuse(self): # fuse model Conv2d() + BatchNorm2d() layers
LOGGER.info('Fusing layers... ')
for m in self.model.modules():
if isinstance(m, (RepConvN)) and hasattr(m, 'fuse_convs'):
m.fuse_convs()
m.forward = m.forward_fuse # update forward
if isinstance(m, (Conv, DWConv)) and hasattr(m, 'bn'):
m.conv = fuse_conv_and_bn(m.conv, m.bn) # update conv
delattr(m, 'bn') # remove batchnorm
m.forward = m.forward_fuse # update forward
self.info()
return self
def info(self, verbose=False, img_size=640): # print model information
model_info(self, verbose, img_size)
def _apply(self, fn):
# Apply to(), cpu(), cuda(), half() to model tensors that are not parameters or registered buffers
self = super()._apply(fn)
m = self.model[-1] # Detect()
if isinstance(m, (Detect, DualDetect, TripleDetect, DDetect, DualDDetect, TripleDDetect, Segment, DSegment, DualDSegment, Panoptic)):
m.stride = fn(m.stride)
m.anchors = fn(m.anchors)
m.strides = fn(m.strides)
# m.grid = list(map(fn, m.grid))
return self
class DetectionModel(BaseModel):
# YOLO detection model
def __init__(self, cfg='yolo.yaml', ch=3, nc=None, anchors=None): # model, input channels, number of classes
super().__init__()
if isinstance(cfg, dict):
self.yaml = cfg # model dict
else: # is *.yaml
import yaml # for torch hub
self.yaml_file = Path(cfg).name
with open(cfg, encoding='ascii', errors='ignore') as f:
self.yaml = yaml.safe_load(f) # model dict
# Define model
ch = self.yaml['ch'] = self.yaml.get('ch', ch) # input channels
if nc and nc != self.yaml['nc']:
LOGGER.info(f"Overriding model.yaml nc={self.yaml['nc']} with nc={nc}")
self.yaml['nc'] = nc # override yaml value
if anchors:
LOGGER.info(f'Overriding model.yaml anchors with anchors={anchors}')
self.yaml['anchors'] = round(anchors) # override yaml value
self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch]) # model, savelist
self.names = [str(i) for i in range(self.yaml['nc'])] # default names
self.inplace = self.yaml.get('inplace', True)
# Build strides, anchors
m = self.model[-1] # Detect()
if isinstance(m, (Detect, DDetect, Segment, DSegment, Panoptic)):
s = 256 # 2x min stride
m.inplace = self.inplace
forward = lambda x: self.forward(x)[0] if isinstance(m, (Segment, DSegment, Panoptic)) else self.forward(x)
m.stride = torch.tensor([s / x.shape[-2] for x in forward(torch.zeros(1, ch, s, s))]) # forward
# check_anchor_order(m)
# m.anchors /= m.stride.view(-1, 1, 1)
self.stride = m.stride
m.bias_init() # only run once
if isinstance(m, (DualDetect, TripleDetect, DualDDetect, TripleDDetect, DualDSegment)):
s = 256 # 2x min stride
m.inplace = self.inplace
forward = lambda x: self.forward(x)[0][0] if isinstance(m, (DualDSegment)) else self.forward(x)[0]
m.stride = torch.tensor([s / x.shape[-2] for x in forward(torch.zeros(1, ch, s, s))]) # forward
# check_anchor_order(m)
# m.anchors /= m.stride.view(-1, 1, 1)
self.stride = m.stride
m.bias_init() # only run once
# Init weights, biases
initialize_weights(self)
self.info()
LOGGER.info('')
def forward(self, x, augment=False, profile=False, visualize=False):
if augment:
return self._forward_augment(x) # augmented inference, None
return self._forward_once(x, profile, visualize) # single-scale inference, train
def _forward_augment(self, x):
img_size = x.shape[-2:] # height, width
s = [1, 0.83, 0.67] # scales
f = [None, 3, None] # flips (2-ud, 3-lr)
y = [] # outputs
for si, fi in zip(s, f):
xi = scale_img(x.flip(fi) if fi else x, si, gs=int(self.stride.max()))
yi = self._forward_once(xi)[0] # forward
# cv2.imwrite(f'img_{si}.jpg', 255 * xi[0].cpu().numpy().transpose((1, 2, 0))[:, :, ::-1]) # save
yi = self._descale_pred(yi, fi, si, img_size)
y.append(yi)
y = self._clip_augmented(y) # clip augmented tails
return torch.cat(y, 1), None # augmented inference, train
def _descale_pred(self, p, flips, scale, img_size):
# de-scale predictions following augmented inference (inverse operation)
if self.inplace:
p[..., :4] /= scale # de-scale
if flips == 2:
p[..., 1] = img_size[0] - p[..., 1] # de-flip ud
elif flips == 3:
p[..., 0] = img_size[1] - p[..., 0] # de-flip lr
else:
x, y, wh = p[..., 0:1] / scale, p[..., 1:2] / scale, p[..., 2:4] / scale # de-scale
if flips == 2:
y = img_size[0] - y # de-flip ud
elif flips == 3:
x = img_size[1] - x # de-flip lr
p = torch.cat((x, y, wh, p[..., 4:]), -1)
return p
def _clip_augmented(self, y):
# Clip YOLO augmented inference tails
nl = self.model[-1].nl # number of detection layers (P3-P5)
g = sum(4 ** x for x in range(nl)) # grid points
e = 1 # exclude layer count
i = (y[0].shape[1] // g) * sum(4 ** x for x in range(e)) # indices
y[0] = y[0][:, :-i] # large
i = (y[-1].shape[1] // g) * sum(4 ** (nl - 1 - x) for x in range(e)) # indices
y[-1] = y[-1][:, i:] # small
return y
Model = DetectionModel # retain YOLO 'Model' class for backwards compatibility
class SegmentationModel(DetectionModel):
# YOLO segmentation model
def __init__(self, cfg='yolo-seg.yaml', ch=3, nc=None, anchors=None):
super().__init__(cfg, ch, nc, anchors)
class ClassificationModel(BaseModel):
# YOLO classification model
def __init__(self, cfg=None, model=None, nc=1000, cutoff=10): # yaml, model, number of classes, cutoff index
super().__init__()
self._from_detection_model(model, nc, cutoff) if model is not None else self._from_yaml(cfg)
def _from_detection_model(self, model, nc=1000, cutoff=10):
# Create a YOLO classification model from a YOLO detection model
if isinstance(model, DetectMultiBackend):
model = model.model # unwrap DetectMultiBackend
model.model = model.model[:cutoff] # backbone
m = model.model[-1] # last layer
ch = m.conv.in_channels if hasattr(m, 'conv') else m.cv1.conv.in_channels # ch into module
c = Classify(ch, nc) # Classify()
c.i, c.f, c.type = m.i, m.f, 'models.common.Classify' # index, from, type
model.model[-1] = c # replace
self.model = model.model
self.stride = model.stride
self.save = []
self.nc = nc
def _from_yaml(self, cfg):
# Create a YOLO classification model from a *.yaml file
self.model = None
def parse_model(d, ch): # model_dict, input_channels(3)
# Parse a YOLO model.yaml dictionary
LOGGER.info(f"\n{'':>3}{'from':>18}{'n':>3}{'params':>10} {'module':<40}{'arguments':<30}")
anchors, nc, gd, gw, act = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple'], d.get('activation')
if act:
Conv.default_act = eval(act) # redefine default activation, i.e. Conv.default_act = nn.SiLU()
RepConvN.default_act = eval(act) # redefine default activation, i.e. Conv.default_act = nn.SiLU()
LOGGER.info(f"{colorstr('activation:')} {act}") # print
na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors # number of anchors
no = na * (nc + 5) # number of outputs = anchors * (classes + 5)
layers, save, c2 = [], [], ch[-1] # layers, savelist, ch out
for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']): # from, number, module, args
m = eval(m) if isinstance(m, str) else m # eval strings
for j, a in enumerate(args):
with contextlib.suppress(NameError):
args[j] = eval(a) if isinstance(a, str) else a # eval strings
n = n_ = max(round(n * gd), 1) if n > 1 else n # depth gain
if m in {
Conv, AConv, ConvTranspose,
Bottleneck, SPP, SPPF, DWConv, BottleneckCSP, nn.ConvTranspose2d, DWConvTranspose2d, SPPCSPC, ADown,
RepNCSPELAN4, SPPELAN}:
c1, c2 = ch[f], args[0]
if c2 != no: # if not output
c2 = make_divisible(c2 * gw, 8)
args = [c1, c2, *args[1:]]
if m in {BottleneckCSP, SPPCSPC}:
args.insert(2, n) # number of repeats
n = 1
elif m is nn.BatchNorm2d:
args = [ch[f]]
elif m is Concat:
c2 = sum(ch[x] for x in f)
elif m is Shortcut:
c2 = ch[f[0]]
elif m is ReOrg:
c2 = ch[f] * 4
elif m is CBLinear:
c2 = args[0]
c1 = ch[f]
args = [c1, c2, *args[1:]]
elif m is CBFuse:
c2 = ch[f[-1]]
# TODO: channel, gw, gd
elif m in {Detect, DualDetect, TripleDetect, DDetect, DualDDetect, TripleDDetect, Segment, DSegment, DualDSegment, Panoptic}:
args.append([ch[x] for x in f])
# if isinstance(args[1], int): # number of anchors
# args[1] = [list(range(args[1] * 2))] * len(f)
if m in {Segment, DSegment, DualDSegment, Panoptic}:
args[2] = make_divisible(args[2] * gw, 8)
elif m is Contract:
c2 = ch[f] * args[0] ** 2
elif m is Expand:
c2 = ch[f] // args[0] ** 2
else:
c2 = ch[f]
m_ = nn.Sequential(*(m(*args) for _ in range(n))) if n > 1 else m(*args) # module
t = str(m)[8:-2].replace('__main__.', '') # module type
np = sum(x.numel() for x in m_.parameters()) # number params
m_.i, m_.f, m_.type, m_.np = i, f, t, np # attach index, 'from' index, type, number params
LOGGER.info(f'{i:>3}{str(f):>18}{n_:>3}{np:10.0f} {t:<40}{str(args):<30}') # print
save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1) # append to savelist
layers.append(m_)
if i == 0:
ch = []
ch.append(c2)
return nn.Sequential(*layers), sorted(save)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--cfg', type=str, default='yolo.yaml', help='model.yaml')
parser.add_argument('--batch-size', type=int, default=1, help='total batch size for all GPUs')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--profile', action='store_true', help='profile model speed')
parser.add_argument('--line-profile', action='store_true', help='profile model speed layer by layer')
parser.add_argument('--test', action='store_true', help='test all yolo*.yaml')
opt = parser.parse_args()
opt.cfg = check_yaml(opt.cfg) # check YAML
print_args(vars(opt))
device = select_device(opt.device)
# Create model
im = torch.rand(opt.batch_size, 3, 640, 640).to(device)
model = Model(opt.cfg).to(device)
model.eval()
# Options
if opt.line_profile: # profile layer by layer
model(im, profile=True)
elif opt.profile: # profile forward-backward
results = profile(input=im, ops=[model], n=3)
elif opt.test: # test all models
for cfg in Path(ROOT / 'models').rglob('yolo*.yaml'):
try:
_ = Model(cfg)
except Exception as e:
print(f'Error in {cfg}: {e}')
else: # report fused model summary
model.fuse()
panoptic/predict.py 0 → 100644
View file @ f969ca34
import argparse
import os
import platform
import sys
from pathlib import Path
import torch
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
from models.common import DetectMultiBackend
from utils.dataloaders import IMG_FORMATS, VID_FORMATS, LoadImages, LoadScreenshots, LoadStreams
from utils.general import (LOGGER, Profile, check_file, check_img_size, check_imshow, check_requirements, colorstr, cv2,
increment_path, non_max_suppression, print_args, scale_boxes, scale_segments,
strip_optimizer, xyxy2xywh)
from utils.plots import Annotator, colors, save_one_box
from utils.segment.general import masks2segments, process_mask
from utils.torch_utils import select_device, smart_inference_mode
@smart_inference_mode()
def run(
weights=ROOT / 'yolo-pan.pt', # model.pt path(s)
source=ROOT / 'data/images', # file/dir/URL/glob/screen/0(webcam)
data=ROOT / 'data/coco128.yaml', # dataset.yaml path
imgsz=(640, 640), # inference size (height, width)
conf_thres=0.25, # confidence threshold
iou_thres=0.45, # NMS IOU threshold
max_det=1000, # maximum detections per image
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
view_img=False, # show results
save_txt=False, # save results to *.txt
save_conf=False, # save confidences in --save-txt labels
save_crop=False, # save cropped prediction boxes
nosave=False, # do not save images/videos
classes=None, # filter by class: --class 0, or --class 0 2 3
agnostic_nms=False, # class-agnostic NMS
augment=False, # augmented inference
visualize=False, # visualize features
update=False, # update all models
project=ROOT / 'runs/predict-seg', # save results to project/name
name='exp', # save results to project/name
exist_ok=False, # existing project/name ok, do not increment
line_thickness=3, # bounding box thickness (pixels)
hide_labels=False, # hide labels
hide_conf=False, # hide confidences
half=False, # use FP16 half-precision inference
dnn=False, # use OpenCV DNN for ONNX inference
vid_stride=1, # video frame-rate stride
retina_masks=False,
):
source = str(source)
save_img = not nosave and not source.endswith('.txt') # save inference images
is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)
is_url = source.lower().startswith(('rtsp://', 'rtmp://', 'http://', 'https://'))
webcam = source.isnumeric() or source.endswith('.txt') or (is_url and not is_file)
screenshot = source.lower().startswith('screen')
if is_url and is_file:
source = check_file(source) # download
# Directories
save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
# Load model
device = select_device(device)
model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
stride, names, pt = model.stride, model.names, model.pt
imgsz = check_img_size(imgsz, s=stride) # check image size
# Dataloader
bs = 1 # batch_size
if webcam:
view_img = check_imshow(warn=True)
dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
bs = len(dataset)
elif screenshot:
dataset = LoadScreenshots(source, img_size=imgsz, stride=stride, auto=pt)
else:
dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
vid_path, vid_writer = [None] * bs, [None] * bs
# Run inference
model.warmup(imgsz=(1 if pt else bs, 3, *imgsz)) # warmup
seen, windows, dt = 0, [], (Profile(), Profile(), Profile())
for path, im, im0s, vid_cap, s in dataset:
with dt[0]:
im = torch.from_numpy(im).to(model.device)
im = im.half() if model.fp16 else im.float() # uint8 to fp16/32
im /= 255 # 0 - 255 to 0.0 - 1.0
if len(im.shape) == 3:
im = im[None] # expand for batch dim
# Inference
with dt[1]:
visualize = increment_path(save_dir / Path(path).stem, mkdir=True) if visualize else False
pred, proto = model(im, augment=augment, visualize=visualize)[:2]
# NMS
with dt[2]:
pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det, nm=32)
# Second-stage classifier (optional)
# pred = utils.general.apply_classifier(pred, classifier_model, im, im0s)
# Process predictions
for i, det in enumerate(pred): # per image
seen += 1
if webcam: # batch_size >= 1
p, im0, frame = path[i], im0s[i].copy(), dataset.count
s += f'{i}: '
else:
p, im0, frame = path, im0s.copy(), getattr(dataset, 'frame', 0)
p = Path(p) # to Path
save_path = str(save_dir / p.name) # im.jpg
txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # im.txt
s += '%gx%g ' % im.shape[2:] # print string
imc = im0.copy() if save_crop else im0 # for save_crop
annotator = Annotator(im0, line_width=line_thickness, example=str(names))
if len(det):
masks = process_mask(proto[i], det[:, 6:], det[:, :4], im.shape[2:], upsample=True) # HWC
det[:, :4] = scale_boxes(im.shape[2:], det[:, :4], im0.shape).round() # rescale boxes to im0 size
# Segments
if save_txt:
segments = reversed(masks2segments(masks))
segments = [scale_segments(im.shape[2:], x, im0.shape, normalize=True) for x in segments]
# Print results
for c in det[:, 5].unique():
n = (det[:, 5] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# Mask plotting
annotator.masks(masks,
colors=[colors(x, True) for x in det[:, 5]],
im_gpu=None if retina_masks else im[i])
# Write results
for j, (*xyxy, conf, cls) in enumerate(reversed(det[:, :6])):
if save_txt: # Write to file
segj = segments[j].reshape(-1) # (n,2) to (n*2)
line = (cls, *segj, conf) if save_conf else (cls, *segj) # label format
with open(f'{txt_path}.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or save_crop or view_img: # Add bbox to image
c = int(cls) # integer class
label = None if hide_labels else (names[c] if hide_conf else f'{names[c]} {conf:.2f}')
annotator.box_label(xyxy, label, color=colors(c, True))
# annotator.draw.polygon(segments[j], outline=colors(c, True), width=3)
if save_crop:
save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True)
# Stream results
im0 = annotator.result()
if view_img:
if platform.system() == 'Linux' and p not in windows:
windows.append(p)
cv2.namedWindow(str(p), cv2.WINDOW_NORMAL | cv2.WINDOW_KEEPRATIO) # allow window resize (Linux)
cv2.resizeWindow(str(p), im0.shape[1], im0.shape[0])
cv2.imshow(str(p), im0)
if cv2.waitKey(1) == ord('q'): # 1 millisecond
exit()
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video' or 'stream'
if vid_path[i] != save_path: # new video
vid_path[i] = save_path
if isinstance(vid_writer[i], cv2.VideoWriter):
vid_writer[i].release() # release previous video writer
if vid_cap: # video
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
else: # stream
fps, w, h = 30, im0.shape[1], im0.shape[0]
save_path = str(Path(save_path).with_suffix('.mp4')) # force *.mp4 suffix on results videos
vid_writer[i] = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))
vid_writer[i].write(im0)
# Print time (inference-only)
LOGGER.info(f"{s}{'' if len(det) else '(no detections), '}{dt[1].dt * 1E3:.1f}ms")
# Print results
t = tuple(x.t / seen * 1E3 for x in dt) # speeds per image
LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}' % t)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
if update:
strip_optimizer(weights[0]) # update model (to fix SourceChangeWarning)
def parse_opt():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolo-pan.pt', help='model path(s)')
parser.add_argument('--source', type=str, default=ROOT / 'data/images', help='file/dir/URL/glob/screen/0(webcam)')
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='(optional) dataset.yaml path')
parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[640], help='inference size h,w')
parser.add_argument('--conf-thres', type=float, default=0.25, help='confidence threshold')
parser.add_argument('--iou-thres', type=float, default=0.45, help='NMS IoU threshold')
parser.add_argument('--max-det', type=int, default=1000, help='maximum detections per image')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--view-img', action='store_true', help='show results')
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes')
parser.add_argument('--nosave', action='store_true', help='do not save images/videos')
parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --classes 0, or --classes 0 2 3')
parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
parser.add_argument('--augment', action='store_true', help='augmented inference')
parser.add_argument('--visualize', action='store_true', help='visualize features')
parser.add_argument('--update', action='store_true', help='update all models')
parser.add_argument('--project', default=ROOT / 'runs/predict-seg', help='save results to project/name')
parser.add_argument('--name', default='exp', help='save results to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--line-thickness', default=3, type=int, help='bounding box thickness (pixels)')
parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels')
parser.add_argument('--hide-conf', default=False, action='store_true', help='hide confidences')
parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')
parser.add_argument('--vid-stride', type=int, default=1, help='video frame-rate stride')
parser.add_argument('--retina-masks', action='store_true', help='whether to plot masks in native resolution')
opt = parser.parse_args()
opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1 # expand
print_args(vars(opt))
return opt
def main(opt):
check_requirements(exclude=('tensorboard', 'thop'))
run(**vars(opt))
if __name__ == "__main__":
opt = parse_opt()
main(opt)
panoptic/train.py 0 → 100644
View file @ f969ca34
import argparse
import math
import os
import random
import sys
import time
from copy import deepcopy
from datetime import datetime
from pathlib import Path
import numpy as np
import torch
import torch.distributed as dist
import torch.nn as nn
import yaml
from torch.optim import lr_scheduler
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
import panoptic.val as validate # for end-of-epoch mAP
from models.experimental import attempt_load
from models.yolo import SegmentationModel
from utils.autoanchor import check_anchors
from utils.autobatch import check_train_batch_size
from utils.callbacks import Callbacks
from utils.downloads import attempt_download, is_url
from utils.general import (LOGGER, TQDM_BAR_FORMAT, check_amp, check_dataset, check_file, check_git_info,
check_git_status, check_img_size, check_requirements, check_suffix, check_yaml, colorstr,
get_latest_run, increment_path, init_seeds, intersect_dicts, labels_to_class_weights,
labels_to_image_weights, one_cycle, one_flat_cycle, print_args, print_mutation, strip_optimizer, yaml_save)
from utils.loggers import GenericLogger
from utils.plots import plot_evolve, plot_labels
from utils.panoptic.dataloaders import create_dataloader
from utils.panoptic.loss_tal import ComputeLoss
from utils.panoptic.metrics import KEYS, fitness
from utils.panoptic.plots import plot_images_and_masks, plot_results_with_masks
from utils.torch_utils import (EarlyStopping, ModelEMA, de_parallel, select_device, smart_DDP, smart_optimizer,
smart_resume, torch_distributed_zero_first)
LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1)) # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
GIT_INFO = None#check_git_info()
def train(hyp, opt, device, callbacks): # hyp is path/to/hyp.yaml or hyp dictionary
save_dir, epochs, batch_size, weights, single_cls, evolve, data, cfg, resume, noval, nosave, workers, freeze, mask_ratio = \
Path(opt.save_dir), opt.epochs, opt.batch_size, opt.weights, opt.single_cls, opt.evolve, opt.data, opt.cfg, \
opt.resume, opt.noval, opt.nosave, opt.workers, opt.freeze, opt.mask_ratio
# callbacks.run('on_pretrain_routine_start')
# Directories
w = save_dir / 'weights' # weights dir
(w.parent if evolve else w).mkdir(parents=True, exist_ok=True) # make dir
last, best = w / 'last.pt', w / 'best.pt'
# Hyperparameters
if isinstance(hyp, str):
with open(hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
LOGGER.info(colorstr('hyperparameters: ') + ', '.join(f'{k}={v}' for k, v in hyp.items()))
opt.hyp = hyp.copy() # for saving hyps to checkpoints
# Save run settings
if not evolve:
yaml_save(save_dir / 'hyp.yaml', hyp)
yaml_save(save_dir / 'opt.yaml', vars(opt))
# Loggers
data_dict = None
if RANK in {-1, 0}:
logger = GenericLogger(opt=opt, console_logger=LOGGER)
# Config
plots = not evolve and not opt.noplots # create plots
overlap = not opt.no_overlap
cuda = device.type != 'cpu'
init_seeds(opt.seed + 1 + RANK, deterministic=True)
with torch_distributed_zero_first(LOCAL_RANK):
data_dict = data_dict or check_dataset(data) # check if None
train_path, val_path = data_dict['train'], data_dict['val']
nc = 1 if single_cls else int(data_dict['nc']) # number of classes
names = {0: 'item'} if single_cls and len(data_dict['names']) != 1 else data_dict['names'] # class names
#is_coco = isinstance(val_path, str) and val_path.endswith('coco/val2017.txt') # COCO dataset
is_coco = isinstance(val_path, str) and val_path.endswith('val2017.txt') # COCO dataset
# Model
check_suffix(weights, '.pt') # check weights
pretrained = weights.endswith('.pt')
if pretrained:
with torch_distributed_zero_first(LOCAL_RANK):
weights = attempt_download(weights) # download if not found locally
ckpt = torch.load(weights, map_location='cpu') # load checkpoint to CPU to avoid CUDA memory leak
model = SegmentationModel(cfg or ckpt['model'].yaml, ch=3, nc=nc).to(device)
exclude = ['anchor'] if (cfg or hyp.get('anchors')) and not resume else [] # exclude keys
csd = ckpt['model'].float().state_dict() # checkpoint state_dict as FP32
csd = intersect_dicts(csd, model.state_dict(), exclude=exclude) # intersect
model.load_state_dict(csd, strict=False) # load
LOGGER.info(f'Transferred {len(csd)}/{len(model.state_dict())} items from {weights}') # report
else:
model = SegmentationModel(cfg, ch=3, nc=nc).to(device) # create
amp = check_amp(model) # check AMP
# Freeze
freeze = [f'model.{x}.' for x in (freeze if len(freeze) > 1 else range(freeze[0]))] # layers to freeze
for k, v in model.named_parameters():
#v.requires_grad = True # train all layers
# v.register_hook(lambda x: torch.nan_to_num(x)) # NaN to 0 (commented for erratic training results)
if any(x in k for x in freeze):
LOGGER.info(f'freezing {k}')
v.requires_grad = False
# Image size
gs = max(int(model.stride.max()), 32) # grid size (max stride)
imgsz = check_img_size(opt.imgsz, gs, floor=gs * 2) # verify imgsz is gs-multiple
# Batch size
if RANK == -1 and batch_size == -1: # single-GPU only, estimate best batch size
batch_size = check_train_batch_size(model, imgsz, amp)
logger.update_params({"batch_size": batch_size})
# loggers.on_params_update({"batch_size": batch_size})
# Optimizer
nbs = 64 # nominal batch size
accumulate = max(round(nbs / batch_size), 1) # accumulate loss before optimizing
hyp['weight_decay'] *= batch_size * accumulate / nbs # scale weight_decay
optimizer = smart_optimizer(model, opt.optimizer, hyp['lr0'], hyp['momentum'], hyp['weight_decay'])
# Scheduler
if opt.cos_lr:
lf = one_cycle(1, hyp['lrf'], epochs) # cosine 1->hyp['lrf']
elif opt.flat_cos_lr:
lf = one_flat_cycle(1, hyp['lrf'], epochs) # cosine 1->hyp['lrf']
elif opt.fixed_lr:
lf = lambda x: 1.0
elif opt.poly_lr:
power = 0.9
lf = lambda x: ((1 - (x / epochs)) ** power) * (1.0 - hyp['lrf']) + hyp['lrf']
else:
lf = lambda x: (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf'] # linear
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf) # plot_lr_scheduler(optimizer, scheduler, epochs)
# EMA
ema = ModelEMA(model) if RANK in {-1, 0} else None
# Resume
best_fitness, start_epoch = 0.0, 0
if pretrained:
if resume:
best_fitness, start_epoch, epochs = smart_resume(ckpt, optimizer, ema, weights, epochs, resume)
del ckpt, csd
# DP mode
if cuda and RANK == -1 and torch.cuda.device_count() > 1:
LOGGER.warning('WARNING ⚠️ DP not recommended, use torch.distributed.run for best DDP Multi-GPU results.')
model = torch.nn.DataParallel(model)
# SyncBatchNorm
if opt.sync_bn and cuda and RANK != -1:
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
LOGGER.info('Using SyncBatchNorm()')
# Trainloader
train_loader, dataset = create_dataloader(
train_path,
imgsz,
batch_size // WORLD_SIZE,
gs,
single_cls,
hyp=hyp,
augment=True,
cache=None if opt.cache == 'val' else opt.cache,
rect=opt.rect,
rank=LOCAL_RANK,
workers=workers,
image_weights=opt.image_weights,
close_mosaic=opt.close_mosaic != 0,
quad=opt.quad,
prefix=colorstr('train: '),
shuffle=True,
mask_downsample_ratio=mask_ratio,
overlap_mask=overlap,
)
labels = np.concatenate(dataset.labels, 0)
mlc = int(labels[:, 0].max()) # max label class
assert mlc < nc, f'Label class {mlc} exceeds nc={nc} in {data}. Possible class labels are 0-{nc - 1}'
# Process 0
if RANK in {-1, 0}:
val_loader = create_dataloader(val_path,
imgsz,
batch_size // WORLD_SIZE * 2,
gs,
single_cls,
hyp=hyp,
cache=None if noval else opt.cache,
rect=True,
rank=-1,
workers=workers * 2,
pad=0.5,
mask_downsample_ratio=mask_ratio,
overlap_mask=overlap,
prefix=colorstr('val: '))[0]
if not resume:
#if not opt.noautoanchor:
# check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz) # run AutoAnchor
model.half().float() # pre-reduce anchor precision
if plots:
plot_labels(labels, names, save_dir)
# callbacks.run('on_pretrain_routine_end', labels, names)
# DDP mode
if cuda and RANK != -1:
model = smart_DDP(model)
# Model attributes
nl = de_parallel(model).model[-1].nl # number of detection layers (to scale hyps)
#hyp['box'] *= 3 / nl # scale to layers
#hyp['cls'] *= nc / 80 * 3 / nl # scale to classes and layers
#hyp['obj'] *= (imgsz / 640) ** 2 * 3 / nl # scale to image size and layers
hyp['label_smoothing'] = opt.label_smoothing
model.nc = nc # attach number of classes to model
model.hyp = hyp # attach hyperparameters to model
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) * nc # attach class weights
model.names = names
# Start training
t0 = time.time()
nb = len(train_loader) # number of batches
nw = max(round(hyp['warmup_epochs'] * nb), 100) # number of warmup iterations, max(3 epochs, 100 iterations)
# nw = min(nw, (epochs - start_epoch) / 2 * nb) # limit warmup to < 1/2 of training
last_opt_step = -1
maps = np.zeros(nc) # mAP per class
results = (0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
scheduler.last_epoch = start_epoch - 1 # do not move
scaler = torch.cuda.amp.GradScaler(enabled=amp)
stopper, stop = EarlyStopping(patience=opt.patience), False
compute_loss = ComputeLoss(model, overlap=overlap) # init loss class
# callbacks.run('on_train_start')
LOGGER.info(f'Image sizes {imgsz} train, {imgsz} val\n'
f'Using {train_loader.num_workers * WORLD_SIZE} dataloader workers\n'
f"Logging results to {colorstr('bold', save_dir)}\n"
f'Starting training for {epochs} epochs...')
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
# callbacks.run('on_train_epoch_start')
model.train()
# Update image weights (optional, single-GPU only)
if opt.image_weights:
cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 / nc # class weights
iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw) # image weights
dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n) # rand weighted idx
if epoch == (epochs - opt.close_mosaic):
LOGGER.info("Closing dataloader mosaic")
dataset.mosaic = False
# Update mosaic border (optional)
# b = int(random.uniform(0.25 * imgsz, 0.75 * imgsz + gs) // gs * gs)
# dataset.mosaic_border = [b - imgsz, -b] # height, width borders
mloss = torch.zeros(6, device=device) # mean losses
if RANK != -1:
train_loader.sampler.set_epoch(epoch)
pbar = enumerate(train_loader)
LOGGER.info(('\n' + '%11s' * 10) %
('Epoch', 'GPU_mem', 'box_loss', 'seg_loss', 'cls_loss', 'dfl_loss', 'fcl_loss', 'dic_loss', 'Instances', 'Size'))
if RANK in {-1, 0}:
pbar = tqdm(pbar, total=nb, bar_format=TQDM_BAR_FORMAT) # progress bar
optimizer.zero_grad()
for i, (imgs, targets, paths, _, masks, semasks) in pbar: # batch ------------------------------------------------------
# callbacks.run('on_train_batch_start')
#print(imgs.shape)
#print(semasks.shape)
#print(masks.shape)
ni = i + nb * epoch # number integrated batches (since train start)
imgs = imgs.to(device, non_blocking=True).float() / 255 # uint8 to float32, 0-255 to 0.0-1.0
# Warmup
if ni <= nw:
xi = [0, nw] # x interp
# compute_loss.gr = np.interp(ni, xi, [0.0, 1.0]) # iou loss ratio (obj_loss = 1.0 or iou)
accumulate = max(1, np.interp(ni, xi, [1, nbs / batch_size]).round())
for j, x in enumerate(optimizer.param_groups):
# bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
x['lr'] = np.interp(ni, xi, [hyp['warmup_bias_lr'] if j == 0 else 0.0, x['initial_lr'] * lf(epoch)])
if 'momentum' in x:
x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])
# Multi-scale
if opt.multi_scale:
sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs # size
sf = sz / max(imgs.shape[2:]) # scale factor
if sf != 1:
ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]] # new shape (stretched to gs-multiple)
imgs = nn.functional.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)
# Forward
with torch.cuda.amp.autocast(amp):
pred = model(imgs) # forward
loss, loss_items = compute_loss(pred, targets.to(device), masks=masks.to(device).float(),
semasks=semasks.to(device).float())
if RANK != -1:
loss *= WORLD_SIZE # gradient averaged between devices in DDP mode
if opt.quad:
loss *= 4.
# Backward
torch.use_deterministic_algorithms(False)
scaler.scale(loss).backward()
# Optimize - https://pytorch.org/docs/master/notes/amp_examples.html
if ni - last_opt_step >= accumulate:
scaler.unscale_(optimizer) # unscale gradients
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=10.0) # clip gradients
scaler.step(optimizer) # optimizer.step
scaler.update()
optimizer.zero_grad()
if ema:
ema.update(model)
last_opt_step = ni
# Log
if RANK in {-1, 0}:
mloss = (mloss * i + loss_items) / (i + 1) # update mean losses
mem = f'{torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0:.3g}G' # (GB)
pbar.set_description(('%11s' * 2 + '%11.4g' * 8) %
(f'{epoch}/{epochs - 1}', mem, *mloss, targets.shape[0], imgs.shape[-1]))
# callbacks.run('on_train_batch_end', model, ni, imgs, targets, paths)
# if callbacks.stop_training:
# return
# Mosaic plots
if plots:
if ni < 10:
plot_images_and_masks(imgs, targets, masks, semasks, paths, save_dir / f"train_batch{ni}.jpg")
if ni == 10:
files = sorted(save_dir.glob('train*.jpg'))
logger.log_images(files, "Mosaics", epoch)
# end batch ------------------------------------------------------------------------------------------------
# Scheduler
lr = [x['lr'] for x in optimizer.param_groups] # for loggers
scheduler.step()
if RANK in {-1, 0}:
# mAP
# callbacks.run('on_train_epoch_end', epoch=epoch)
ema.update_attr(model, include=['yaml', 'nc', 'hyp', 'names', 'stride', 'class_weights'])
final_epoch = (epoch + 1 == epochs) or stopper.possible_stop
if not noval or final_epoch: # Calculate mAP
if (opt.save_period > 0 and epoch % opt.save_period == 0) or (epoch > (epochs - 2 * opt.close_mosaic)):
results, maps, _ = validate.run(data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
half=amp,
model=ema.ema,
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
plots=False,
callbacks=callbacks,
compute_loss=compute_loss,
mask_downsample_ratio=mask_ratio,
overlap=overlap)
# Update best mAP
fi = fitness(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
stop = stopper(epoch=epoch, fitness=fi) # early stop check
if fi > best_fitness:
best_fitness = fi
log_vals = list(mloss) + list(results) + lr
# callbacks.run('on_fit_epoch_end', log_vals, epoch, best_fitness, fi)
# Log val metrics and media
metrics_dict = dict(zip(KEYS, log_vals))
logger.log_metrics(metrics_dict, epoch)
# Save model
if (not nosave) or (final_epoch and not evolve): # if save
ckpt = {
'epoch': epoch,
'best_fitness': best_fitness,
'model': deepcopy(de_parallel(model)).half(),
'ema': deepcopy(ema.ema).half(),
'updates': ema.updates,
'optimizer': optimizer.state_dict(),
'opt': vars(opt),
'git': GIT_INFO, # {remote, branch, commit} if a git repo
'date': datetime.now().isoformat()}
# Save last, best and delete
torch.save(ckpt, last)
if best_fitness == fi:
torch.save(ckpt, best)
if opt.save_period > 0 and epoch % opt.save_period == 0:
torch.save(ckpt, w / f'epoch{epoch}.pt')
logger.log_model(w / f'epoch{epoch}.pt')
del ckpt
# callbacks.run('on_model_save', last, epoch, final_epoch, best_fitness, fi)
# EarlyStopping
if RANK != -1: # if DDP training
broadcast_list = [stop if RANK == 0 else None]
dist.broadcast_object_list(broadcast_list, 0) # broadcast 'stop' to all ranks
if RANK != 0:
stop = broadcast_list[0]
if stop:
break # must break all DDP ranks
# end epoch ----------------------------------------------------------------------------------------------------
# end training -----------------------------------------------------------------------------------------------------
if RANK in {-1, 0}:
LOGGER.info(f'\n{epoch - start_epoch + 1} epochs completed in {(time.time() - t0) / 3600:.3f} hours.')
for f in last, best:
if f.exists():
strip_optimizer(f) # strip optimizers
if f is best:
LOGGER.info(f'\nValidating {f}...')
results, _, _ = validate.run(
data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
model=attempt_load(f, device).half(),
iou_thres=0.65 if is_coco else 0.60, # best pycocotools at iou 0.65
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
save_json=is_coco,
verbose=True,
plots=plots,
callbacks=callbacks,
compute_loss=compute_loss,
mask_downsample_ratio=mask_ratio,
overlap=overlap) # val best model with plots
if is_coco:
# callbacks.run('on_fit_epoch_end', list(mloss) + list(results) + lr, epoch, best_fitness, fi)
metrics_dict = dict(zip(KEYS, list(mloss) + list(results) + lr))
logger.log_metrics(metrics_dict, epoch)
# callbacks.run('on_train_end', last, best, epoch, results)
# on train end callback using genericLogger
logger.log_metrics(dict(zip(KEYS[6:22], results)), epochs)
if not opt.evolve:
logger.log_model(best, epoch)
if plots:
plot_results_with_masks(file=save_dir / 'results.csv') # save results.png
files = ['results.png', 'confusion_matrix.png', *(f'{x}_curve.png' for x in ('F1', 'PR', 'P', 'R'))]
files = [(save_dir / f) for f in files if (save_dir / f).exists()] # filter
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}")
logger.log_images(files, "Results", epoch + 1)
logger.log_images(sorted(save_dir.glob('val*.jpg')), "Validation", epoch + 1)
torch.cuda.empty_cache()
return results
def parse_opt(known=False):
parser = argparse.ArgumentParser()
parser.add_argument('--weights', type=str, default=ROOT / 'yolo-pan.pt', help='initial weights path')
parser.add_argument('--cfg', type=str, default='', help='model.yaml path')
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128-seg.yaml', help='dataset.yaml path')
parser.add_argument('--hyp', type=str, default=ROOT / 'data/hyps/hyp.scratch-low.yaml', help='hyperparameters path')
parser.add_argument('--epochs', type=int, default=100, help='total training epochs')
parser.add_argument('--batch-size', type=int, default=16, help='total batch size for all GPUs, -1 for autobatch')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='train, val image size (pixels)')
parser.add_argument('--rect', action='store_true', help='rectangular training')
parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
parser.add_argument('--noval', action='store_true', help='only validate final epoch')
parser.add_argument('--noautoanchor', action='store_true', help='disable AutoAnchor')
parser.add_argument('--noplots', action='store_true', help='save no plot files')
parser.add_argument('--evolve', type=int, nargs='?', const=300, help='evolve hyperparameters for x generations')
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
parser.add_argument('--cache', type=str, nargs='?', const='ram', help='image --cache ram/disk')
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
parser.add_argument('--single-cls', action='store_true', help='train multi-class data as single-class')
parser.add_argument('--optimizer', type=str, choices=['SGD', 'Adam', 'AdamW', 'LION'], default='SGD', help='optimizer')
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
parser.add_argument('--project', default=ROOT / 'runs/train-pan', help='save to project/name')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--quad', action='store_true', help='quad dataloader')
parser.add_argument('--cos-lr', action='store_true', help='cosine LR scheduler')
parser.add_argument('--flat-cos-lr', action='store_true', help='cosine LR scheduler')
parser.add_argument('--fixed-lr', action='store_true', help='fixed LR scheduler')
parser.add_argument('--poly-lr', action='store_true', help='fixed LR scheduler')
parser.add_argument('--label-smoothing', type=float, default=0.0, help='Label smoothing epsilon')
parser.add_argument('--patience', type=int, default=100, help='EarlyStopping patience (epochs without improvement)')
parser.add_argument('--freeze', nargs='+', type=int, default=[0], help='Freeze layers: backbone=10, first3=0 1 2')
parser.add_argument('--save-period', type=int, default=-1, help='Save checkpoint every x epochs (disabled if < 1)')
parser.add_argument('--seed', type=int, default=0, help='Global training seed')
parser.add_argument('--local_rank', type=int, default=-1, help='Automatic DDP Multi-GPU argument, do not modify')
parser.add_argument('--close-mosaic', type=int, default=0, help='Experimental')
# Instance Segmentation Args
parser.add_argument('--mask-ratio', type=int, default=4, help='Downsample the truth masks to saving memory')
parser.add_argument('--no-overlap', action='store_true', help='Overlap masks train faster at slightly less mAP')
return parser.parse_known_args()[0] if known else parser.parse_args()
def main(opt, callbacks=Callbacks()):
# Checks
if RANK in {-1, 0}:
print_args(vars(opt))
#check_git_status()
#check_requirements()
# Resume
if opt.resume and not opt.evolve: # resume from specified or most recent last.pt
last = Path(check_file(opt.resume) if isinstance(opt.resume, str) else get_latest_run())
opt_yaml = last.parent.parent / 'opt.yaml' # train options yaml
opt_data = opt.data # original dataset
if opt_yaml.is_file():
with open(opt_yaml, errors='ignore') as f:
d = yaml.safe_load(f)
else:
d = torch.load(last, map_location='cpu')['opt']
opt = argparse.Namespace(**d) # replace
opt.cfg, opt.weights, opt.resume = '', str(last), True # reinstate
if is_url(opt_data):
opt.data = check_file(opt_data) # avoid HUB resume auth timeout
else:
opt.data, opt.cfg, opt.hyp, opt.weights, opt.project = \
check_file(opt.data), check_yaml(opt.cfg), check_yaml(opt.hyp), str(opt.weights), str(opt.project) # checks
assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
if opt.evolve:
if opt.project == str(ROOT / 'runs/train'): # if default project name, rename to runs/evolve
opt.project = str(ROOT / 'runs/evolve')
opt.exist_ok, opt.resume = opt.resume, False # pass resume to exist_ok and disable resume
if opt.name == 'cfg':
opt.name = Path(opt.cfg).stem # use model.yaml as name
opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok))
# DDP mode
device = select_device(opt.device, batch_size=opt.batch_size)
if LOCAL_RANK != -1:
msg = 'is not compatible with YOLO Multi-GPU DDP training'
assert not opt.image_weights, f'--image-weights {msg}'
assert not opt.evolve, f'--evolve {msg}'
assert opt.batch_size != -1, f'AutoBatch with --batch-size -1 {msg}, please pass a valid --batch-size'
assert opt.batch_size % WORLD_SIZE == 0, f'--batch-size {opt.batch_size} must be multiple of WORLD_SIZE'
assert torch.cuda.device_count() > LOCAL_RANK, 'insufficient CUDA devices for DDP command'
torch.cuda.set_device(LOCAL_RANK)
device = torch.device('cuda', LOCAL_RANK)
dist.init_process_group(backend="nccl" if dist.is_nccl_available() else "gloo")
# Train
if not opt.evolve:
train(opt.hyp, opt, device, callbacks)
# Evolve hyperparameters (optional)
else:
# Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
meta = {
'lr0': (1, 1e-5, 1e-1), # initial learning rate (SGD=1E-2, Adam=1E-3)
'lrf': (1, 0.01, 1.0), # final OneCycleLR learning rate (lr0 * lrf)
'momentum': (0.3, 0.6, 0.98), # SGD momentum/Adam beta1
'weight_decay': (1, 0.0, 0.001), # optimizer weight decay
'warmup_epochs': (1, 0.0, 5.0), # warmup epochs (fractions ok)
'warmup_momentum': (1, 0.0, 0.95), # warmup initial momentum
'warmup_bias_lr': (1, 0.0, 0.2), # warmup initial bias lr
'box': (1, 0.02, 0.2), # box loss gain
'cls': (1, 0.2, 4.0), # cls loss gain
'cls_pw': (1, 0.5, 2.0), # cls BCELoss positive_weight
'obj': (1, 0.2, 4.0), # obj loss gain (scale with pixels)
'obj_pw': (1, 0.5, 2.0), # obj BCELoss positive_weight
'iou_t': (0, 0.1, 0.7), # IoU training threshold
'anchor_t': (1, 2.0, 8.0), # anchor-multiple threshold
'anchors': (2, 2.0, 10.0), # anchors per output grid (0 to ignore)
'fl_gamma': (0, 0.0, 2.0), # focal loss gamma (efficientDet default gamma=1.5)
'hsv_h': (1, 0.0, 0.1), # image HSV-Hue augmentation (fraction)
'hsv_s': (1, 0.0, 0.9), # image HSV-Saturation augmentation (fraction)
'hsv_v': (1, 0.0, 0.9), # image HSV-Value augmentation (fraction)
'degrees': (1, 0.0, 45.0), # image rotation (+/- deg)
'translate': (1, 0.0, 0.9), # image translation (+/- fraction)
'scale': (1, 0.0, 0.9), # image scale (+/- gain)
'shear': (1, 0.0, 10.0), # image shear (+/- deg)
'perspective': (0, 0.0, 0.001), # image perspective (+/- fraction), range 0-0.001
'flipud': (1, 0.0, 1.0), # image flip up-down (probability)
'fliplr': (0, 0.0, 1.0), # image flip left-right (probability)
'mosaic': (1, 0.0, 1.0), # image mixup (probability)
'mixup': (1, 0.0, 1.0), # image mixup (probability)
'copy_paste': (1, 0.0, 1.0)} # segment copy-paste (probability)
with open(opt.hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
if 'anchors' not in hyp: # anchors commented in hyp.yaml
hyp['anchors'] = 3
if opt.noautoanchor:
del hyp['anchors'], meta['anchors']
opt.noval, opt.nosave, save_dir = True, True, Path(opt.save_dir) # only val/save final epoch
# ei = [isinstance(x, (int, float)) for x in hyp.values()] # evolvable indices
evolve_yaml, evolve_csv = save_dir / 'hyp_evolve.yaml', save_dir / 'evolve.csv'
if opt.bucket:
os.system(f'gsutil cp gs://{opt.bucket}/evolve.csv {evolve_csv}') # download evolve.csv if exists
for _ in range(opt.evolve): # generations to evolve
if evolve_csv.exists(): # if evolve.csv exists: select best hyps and mutate
# Select parent(s)
parent = 'single' # parent selection method: 'single' or 'weighted'
x = np.loadtxt(evolve_csv, ndmin=2, delimiter=',', skiprows=1)
n = min(5, len(x)) # number of previous results to consider
x = x[np.argsort(-fitness(x))][:n] # top n mutations
w = fitness(x) - fitness(x).min() + 1E-6 # weights (sum > 0)
if parent == 'single' or len(x) == 1:
# x = x[random.randint(0, n - 1)] # random selection
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
elif parent == 'weighted':
x = (x * w.reshape(n, 1)).sum(0) / w.sum() # weighted combination
# Mutate
mp, s = 0.8, 0.2 # mutation probability, sigma
npr = np.random
npr.seed(int(time.time()))
g = np.array([meta[k][0] for k in hyp.keys()]) # gains 0-1
ng = len(meta)
v = np.ones(ng)
while all(v == 1): # mutate until a change occurs (prevent duplicates)
v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
for i, k in enumerate(hyp.keys()): # plt.hist(v.ravel(), 300)
hyp[k] = float(x[i + 7] * v[i]) # mutate
# Constrain to limits
for k, v in meta.items():
hyp[k] = max(hyp[k], v[1]) # lower limit
hyp[k] = min(hyp[k], v[2]) # upper limit
hyp[k] = round(hyp[k], 5) # significant digits
# Train mutation
results = train(hyp.copy(), opt, device, callbacks)
callbacks = Callbacks()
# Write mutation results
print_mutation(KEYS, results, hyp.copy(), save_dir, opt.bucket)
# Plot results
plot_evolve(evolve_csv)
LOGGER.info(f'Hyperparameter evolution finished {opt.evolve} generations\n'
f"Results saved to {colorstr('bold', save_dir)}\n"
f'Usage example: $ python train.py --hyp {evolve_yaml}')
def run(**kwargs):
# Usage: import train; train.run(data='coco128.yaml', imgsz=320, weights='yolo.pt')
opt = parse_opt(True)
for k, v in kwargs.items():
setattr(opt, k, v)
main(opt)
return opt
if __name__ == "__main__":
opt = parse_opt()
main(opt)
panoptic/val.py 0 → 100644
View file @ f969ca34
import argparse
import json
import os
import sys
from multiprocessing.pool import ThreadPool
from pathlib import Path
import numpy as np
import torch
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
import torch.nn.functional as F
import torchvision.transforms as transforms
from pycocotools import mask as maskUtils
from models.common import DetectMultiBackend
from models.yolo import SegmentationModel
from utils.callbacks import Callbacks
from utils.coco_utils import getCocoIds, getMappingId, getMappingIndex
from utils.general import (LOGGER, NUM_THREADS, TQDM_BAR_FORMAT, Profile, check_dataset, check_img_size,
check_requirements, check_yaml, coco80_to_coco91_class, colorstr, increment_path,
non_max_suppression, print_args, scale_boxes, xywh2xyxy, xyxy2xywh)
from utils.metrics import ConfusionMatrix, box_iou
from utils.plots import output_to_target, plot_val_study
from utils.panoptic.dataloaders import create_dataloader
from utils.panoptic.general import mask_iou, process_mask, process_mask_upsample, scale_image
from utils.panoptic.metrics import Metrics, ap_per_class_box_and_mask, Semantic_Metrics
from utils.panoptic.plots import plot_images_and_masks
from utils.torch_utils import de_parallel, select_device, smart_inference_mode
def save_one_txt(predn, save_conf, shape, file):
# Save one txt result
gn = torch.tensor(shape)[[1, 0, 1, 0]] # normalization gain whwh
for *xyxy, conf, cls in predn.tolist():
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format
with open(file, 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
def save_one_json(predn, jdict, path, class_map, pred_masks):
# Save one JSON result {"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236}
from pycocotools.mask import encode
def single_encode(x):
rle = encode(np.asarray(x[:, :, None], order="F", dtype="uint8"))[0]
rle["counts"] = rle["counts"].decode("utf-8")
return rle
image_id = int(path.stem) if path.stem.isnumeric() else path.stem
box = xyxy2xywh(predn[:, :4]) # xywh
box[:, :2] -= box[:, 2:] / 2 # xy center to top-left corner
pred_masks = np.transpose(pred_masks, (2, 0, 1))
with ThreadPool(NUM_THREADS) as pool:
rles = pool.map(single_encode, pred_masks)
for i, (p, b) in enumerate(zip(predn.tolist(), box.tolist())):
jdict.append({
'image_id': image_id,
'category_id': class_map[int(p[5])],
'bbox': [round(x, 3) for x in b],
'score': round(p[4], 5),
'segmentation': rles[i]})
def process_batch(detections, labels, iouv, pred_masks=None, gt_masks=None, overlap=False, masks=False):
"""
Return correct prediction matrix
Arguments:
detections (array[N, 6]), x1, y1, x2, y2, conf, class
labels (array[M, 5]), class, x1, y1, x2, y2
Returns:
correct (array[N, 10]), for 10 IoU levels
"""
if masks:
if overlap:
nl = len(labels)
index = torch.arange(nl, device=gt_masks.device).view(nl, 1, 1) + 1
gt_masks = gt_masks.repeat(nl, 1, 1) # shape(1,640,640) -> (n,640,640)
gt_masks = torch.where(gt_masks == index, 1.0, 0.0)
if gt_masks.shape[1:] != pred_masks.shape[1:]:
gt_masks = F.interpolate(gt_masks[None], pred_masks.shape[1:], mode="bilinear", align_corners=False)[0]
gt_masks = gt_masks.gt_(0.5)
iou = mask_iou(gt_masks.view(gt_masks.shape[0], -1), pred_masks.view(pred_masks.shape[0], -1))
else: # boxes
iou = box_iou(labels[:, 1:], detections[:, :4])
correct = np.zeros((detections.shape[0], iouv.shape[0])).astype(bool)
correct_class = labels[:, 0:1] == detections[:, 5]
for i in range(len(iouv)):
x = torch.where((iou >= iouv[i]) & correct_class) # IoU > threshold and classes match
if x[0].shape[0]:
matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy() # [label, detect, iou]
if x[0].shape[0] > 1:
matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
# matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
correct[matches[:, 1].astype(int), i] = True
return torch.tensor(correct, dtype=torch.bool, device=iouv.device)
@smart_inference_mode()
def run(
data,
weights=None, # model.pt path(s)
batch_size=32, # batch size
imgsz=640, # inference size (pixels)
conf_thres=0.001, # confidence threshold
iou_thres=0.6, # NMS IoU threshold
max_det=300, # maximum detections per image
task='val', # train, val, test, speed or study
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
workers=8, # max dataloader workers (per RANK in DDP mode)
single_cls=False, # treat as single-class dataset
augment=False, # augmented inference
verbose=False, # verbose output
save_txt=False, # save results to *.txt
save_hybrid=False, # save label+prediction hybrid results to *.txt
save_conf=False, # save confidences in --save-txt labels
save_json=False, # save a COCO-JSON results file
project=ROOT / 'runs/val-pan', # save to project/name
name='exp', # save to project/name
exist_ok=False, # existing project/name ok, do not increment
half=True, # use FP16 half-precision inference
dnn=False, # use OpenCV DNN for ONNX inference
model=None,
dataloader=None,
save_dir=Path(''),
plots=True,
overlap=False,
mask_downsample_ratio=1,
compute_loss=None,
callbacks=Callbacks(),
):
if save_json:
check_requirements(['pycocotools'])
process = process_mask_upsample # more accurate
else:
process = process_mask # faster
# Initialize/load model and set device
training = model is not None
if training: # called by train.py
device, pt, jit, engine = next(model.parameters()).device, True, False, False # get model device, PyTorch model
half &= device.type != 'cpu' # half precision only supported on CUDA
model.half() if half else model.float()
nm = de_parallel(model).model[-1].nm # number of masks
else: # called directly
device = select_device(device, batch_size=batch_size)
# Directories
save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
# Load model
model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
stride, pt, jit, engine = model.stride, model.pt, model.jit, model.engine
imgsz = check_img_size(imgsz, s=stride) # check image size
half = model.fp16 # FP16 supported on limited backends with CUDA
nm = de_parallel(model).model.model[-1].nm if isinstance(model, SegmentationModel) else 32 # number of masks
if engine:
batch_size = model.batch_size
else:
device = model.device
if not (pt or jit):
batch_size = 1 # export.py models default to batch-size 1
LOGGER.info(f'Forcing --batch-size 1 square inference (1,3,{imgsz},{imgsz}) for non-PyTorch models')
# Data
data = check_dataset(data) # check
# Configure
model.eval()
cuda = device.type != 'cpu'
#is_coco = isinstance(data.get('val'), str) and data['val'].endswith(f'coco{os.sep}val2017.txt') # COCO dataset
is_coco = isinstance(data.get('val'), str) and data['val'].endswith(f'val2017.txt') # COCO dataset
nc = 1 if single_cls else int(data['nc']) # number of classes
stuff_names = data.get('stuff_names', []) # names of stuff classes
stuff_nc = len(stuff_names) # number of stuff classes
iouv = torch.linspace(0.5, 0.95, 10, device=device) # iou vector for mAP@0.5:0.95
niou = iouv.numel()
# Semantic Segmentation
img_id_list = []
# Dataloader
if not training:
if pt and not single_cls: # check --weights are trained on --data
ncm = model.model.nc
assert ncm == nc, f'{weights} ({ncm} classes) trained on different --data than what you passed ({nc} ' \
f'classes). Pass correct combination of --weights and --data that are trained together.'
model.warmup(imgsz=(1 if pt else batch_size, 3, imgsz, imgsz)) # warmup
pad, rect = (0.0, False) if task == 'speed' else (0.5, pt) # square inference for benchmarks
task = task if task in ('train', 'val', 'test') else 'val' # path to train/val/test images
dataloader = create_dataloader(data[task],
imgsz,
batch_size,
stride,
single_cls,
pad=pad,
rect=rect,
workers=workers,
prefix=colorstr(f'{task}: '),
overlap_mask=overlap,
mask_downsample_ratio=mask_downsample_ratio)[0]
seen = 0
confusion_matrix = ConfusionMatrix(nc=nc)
names = model.names if hasattr(model, 'names') else model.module.names # get class names
if isinstance(names, (list, tuple)): # old format
names = dict(enumerate(names))
class_map = coco80_to_coco91_class() if is_coco else list(range(1000))
s = ('%22s' + '%11s' * 12) % ('Class', 'Images', 'Instances', 'Box(P', "R", "mAP50", "mAP50-95)", "Mask(P", "R",
"mAP50", "mAP50-95)", 'S(MIoU', 'FWIoU)')
dt = Profile(), Profile(), Profile()
metrics = Metrics()
semantic_metrics = Semantic_Metrics(nc = (nc + stuff_nc), device = device)
loss = torch.zeros(6, device=device)
jdict, stats = [], []
semantic_jdict = []
# callbacks.run('on_val_start')
pbar = tqdm(dataloader, desc=s, bar_format=TQDM_BAR_FORMAT) # progress bar
for batch_i, (im, targets, paths, shapes, masks, semasks) in enumerate(pbar):
# callbacks.run('on_val_batch_start')
with dt[0]:
if cuda:
im = im.to(device, non_blocking=True)
targets = targets.to(device)
masks = masks.to(device)
semasks = semasks.to(device)
masks = masks.float()
semasks = semasks.float()
im = im.half() if half else im.float() # uint8 to fp16/32
im /= 255 # 0 - 255 to 0.0 - 1.0
nb, _, height, width = im.shape # batch size, channels, height, width
# Inference
with dt[1]:
preds, train_out = model(im)# if compute_loss else (*model(im, augment=augment)[:2], None)
#train_out, preds, protos = p if len(p) == 3 else p[1]
#preds = p
#train_out = p[1][0] if len(p[1]) == 3 else p[0]
# protos = train_out[-1]
#print(preds.shape)
#print(train_out[0].shape)
#print(train_out[1].shape)
#print(train_out[2].shape)
_, pred_masks, protos, psemasks = train_out
# Loss
if compute_loss:
loss += compute_loss(train_out, targets, masks, semasks = semasks)[1] # box, obj, cls
# NMS
targets[:, 2:] *= torch.tensor((width, height, width, height), device=device) # to pixels
lb = [targets[targets[:, 0] == i, 1:] for i in range(nb)] if save_hybrid else [] # for autolabelling
with dt[2]:
preds = non_max_suppression(preds,
conf_thres,
iou_thres,
labels=lb,
multi_label=True,
agnostic=single_cls,
max_det=max_det,
nm=nm)
# Metrics
plot_masks = [] # masks for plotting
plot_semasks = [] # masks for plotting
if training:
semantic_metrics.update(psemasks, semasks)
else:
_, _, smh, smw = semasks.shape
semantic_metrics.update(torch.nn.functional.interpolate(psemasks, size = (smh, smw), mode = 'bilinear', align_corners = False), semasks)
if plots and batch_i < 3:
plot_semasks.append(psemasks.clone().detach().cpu())
for si, (pred, proto, psemask) in enumerate(zip(preds, protos, psemasks)):
labels = targets[targets[:, 0] == si, 1:]
nl, npr = labels.shape[0], pred.shape[0] # number of labels, predictions
path, shape = Path(paths[si]), shapes[si][0]
image_id = path.stem
img_id_list.append(image_id)
correct_masks = torch.zeros(npr, niou, dtype=torch.bool, device=device) # init
correct_bboxes = torch.zeros(npr, niou, dtype=torch.bool, device=device) # init
seen += 1
if npr == 0:
if nl:
stats.append((correct_masks, correct_bboxes, *torch.zeros((2, 0), device=device), labels[:, 0]))
if plots:
confusion_matrix.process_batch(detections=None, labels=labels[:, 0])
else:
# Masks
midx = [si] if overlap else targets[:, 0] == si
gt_masks = masks[midx]
pred_masks = process(proto, pred[:, 6:], pred[:, :4], shape=im[si].shape[1:])
# Predictions
if single_cls:
pred[:, 5] = 0
predn = pred.clone()
scale_boxes(im[si].shape[1:], predn[:, :4], shape, shapes[si][1]) # native-space pred
# Evaluate
if nl:
tbox = xywh2xyxy(labels[:, 1:5]) # target boxes
scale_boxes(im[si].shape[1:], tbox, shape, shapes[si][1]) # native-space labels
labelsn = torch.cat((labels[:, 0:1], tbox), 1) # native-space labels
correct_bboxes = process_batch(predn, labelsn, iouv)
correct_masks = process_batch(predn, labelsn, iouv, pred_masks, gt_masks, overlap=overlap, masks=True)
if plots:
confusion_matrix.process_batch(predn, labelsn)
stats.append((correct_masks, correct_bboxes, pred[:, 4], pred[:, 5], labels[:, 0])) # (conf, pcls, tcls)
pred_masks = torch.as_tensor(pred_masks, dtype=torch.uint8)
if plots and batch_i < 3:
plot_masks.append(pred_masks[:15].cpu()) # filter top 15 to plot
# Save/log
if save_txt:
save_one_txt(predn, save_conf, shape, file=save_dir / 'labels' / f'{path.stem}.txt')
if save_json:
pred_masks = scale_image(im[si].shape[1:],
pred_masks.permute(1, 2, 0).contiguous().cpu().numpy(), shape, shapes[si][1])
save_one_json(predn, jdict, path, class_map, pred_masks) # append to COCO-JSON dictionary
# callbacks.run('on_val_image_end', pred, predn, path, names, im[si])
# Semantic Segmentation
h0, w0 = shape
# resize
_, mask_h, mask_w = psemask.shape
h_ratio = mask_h / h0
w_ratio = mask_w / w0
if h_ratio == w_ratio:
psemask = torch.nn.functional.interpolate(psemask[None, :], size = (h0, w0), mode = 'bilinear', align_corners = False)
else:
transform = transforms.CenterCrop((h0, w0))
if (1 != h_ratio) and (1 != w_ratio):
h_new = h0 if (h_ratio < w_ratio) else int(mask_h / w_ratio)
w_new = w0 if (h_ratio > w_ratio) else int(mask_w / h_ratio)
psemask = torch.nn.functional.interpolate(psemask[None, :], size = (h_new, w_new), mode = 'bilinear', align_corners = False)
psemask = transform(psemask)
psemask = torch.squeeze(psemask)
nc, h, w = psemask.shape
semantic_mask = torch.flatten(psemask, start_dim = 1).permute(1, 0) # class x h x w -> (h x w) x class
max_idx = semantic_mask.argmax(1)
output_masks = torch.zeros(semantic_mask.shape).scatter(1, max_idx.cpu().unsqueeze(1), 1.0) # one hot: (h x w) x class
output_masks = torch.reshape(output_masks.permute(1, 0), (nc, h, w)) # (h x w) x class -> class x h x w
psemask = output_masks.to(device = device)
# TODO: check is_coco
instances_ids = getCocoIds(name = 'instances')
stuff_mask = torch.zeros((h, w), device = device)
check_semantic_mask = False
for idx, pred_semantic_mask in enumerate(psemask):
category_id = int(getMappingId(idx))
if 183 == category_id:
# set all non-stuff pixels to other
pred_semantic_mask = (torch.logical_xor(stuff_mask, torch.ones((h, w), device = device))).int()
# ignore the classes which all zeros / unlabeled class
if (0 >= torch.max(pred_semantic_mask)) or (0 >= category_id):
continue
if category_id not in instances_ids:
# record all stuff mask
stuff_mask = torch.logical_or(stuff_mask, pred_semantic_mask)
if (category_id not in instances_ids):
rle = maskUtils.encode(np.asfortranarray(pred_semantic_mask.cpu(), dtype = np.uint8))
rle['counts'] = rle['counts'].decode('utf-8')
temp_d = {
'image_id': int(image_id) if image_id.isnumeric() else image_id,
'category_id': category_id,
'segmentation': rle,
'score': 1
}
semantic_jdict.append(temp_d)
check_semantic_mask = True
if not check_semantic_mask:
# append a other mask for evaluation if the image without any mask
other_mask = (torch.ones((h, w), device = device)).int()
rle = maskUtils.encode(np.asfortranarray(other_mask.cpu(), dtype = np.uint8))
rle['counts'] = rle['counts'].decode('utf-8')
temp_d = {
'image_id': int(image_id) if image_id.isnumeric() else image_id,
'category_id': 183,
'segmentation': rle,
'score': 1
}
semantic_jdict.append(temp_d)
# Plot images
if plots and batch_i < 3:
if len(plot_masks):
plot_masks = torch.cat(plot_masks, dim=0)
if len(plot_semasks):
plot_semasks = torch.cat(plot_semasks, dim = 0)
plot_images_and_masks(im, targets, masks, semasks, paths, save_dir / f'val_batch{batch_i}_labels.jpg', names)
plot_images_and_masks(im, output_to_target(preds, max_det=15), plot_masks, plot_semasks, paths,
save_dir / f'val_batch{batch_i}_pred.jpg', names) # pred
# callbacks.run('on_val_batch_end')
# Compute metrics
stats = [torch.cat(x, 0).cpu().numpy() for x in zip(*stats)] # to numpy
if len(stats) and stats[0].any():
results = ap_per_class_box_and_mask(*stats, plot=plots, save_dir=save_dir, names=names)
metrics.update(results)
nt = np.bincount(stats[4].astype(int), minlength=nc) # number of targets per class
# Print results
pf = '%22s' + '%11i' * 2 + '%11.3g' * 10 # print format
LOGGER.info(pf % ("all", seen, nt.sum(), *metrics.mean_results(), *semantic_metrics.results()))
if nt.sum() == 0:
LOGGER.warning(f'WARNING ⚠️ no labels found in {task} set, can not compute metrics without labels')
# Print results per class
if (verbose or (nc < 50 and not training)) and nc > 1 and len(stats):
for i, c in enumerate(metrics.ap_class_index):
LOGGER.info(pf % (names[c], seen, nt[c], *metrics.class_result(i), *semantic_metrics.results()))
# Print speeds
t = tuple(x.t / seen * 1E3 for x in dt) # speeds per image
if not training:
shape = (batch_size, 3, imgsz, imgsz)
LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {shape}' % t)
# Plots
if plots:
confusion_matrix.plot(save_dir=save_dir, names=list(names.values()))
# callbacks.run('on_val_end')
mp_bbox, mr_bbox, map50_bbox, map_bbox, mp_mask, mr_mask, map50_mask, map_mask = metrics.mean_results()
miou_sem, fwiou_sem = semantic_metrics.results()
semantic_metrics.reset()
# Save JSON
if save_json and len(jdict):
w = Path(weights[0] if isinstance(weights, list) else weights).stem if weights is not None else '' # weights
anno_path = Path(data.get('path', '../coco'))
anno_json = str(anno_path / 'annotations/instances_val2017.json') # annotations json
pred_json = str(save_dir / f"{w}_predictions.json") # predictions json
LOGGER.info(f'\nEvaluating pycocotools mAP... saving {pred_json}...')
with open(pred_json, 'w') as f:
json.dump(jdict, f)
semantic_anno_json = str(anno_path / 'annotations/stuff_val2017.json') # annotations json
semantic_pred_json = str(save_dir / f"{w}_predictions_stuff.json") # predictions json
LOGGER.info(f'\nsaving {semantic_pred_json}...')
with open(semantic_pred_json, 'w') as f:
json.dump(semantic_jdict, f)
try: # https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval
anno = COCO(anno_json) # init annotations api
pred = anno.loadRes(pred_json) # init predictions api
results = []
for eval in COCOeval(anno, pred, 'bbox'), COCOeval(anno, pred, 'segm'):
if is_coco:
eval.params.imgIds = [int(Path(x).stem) for x in dataloader.dataset.im_files] # img ID to evaluate
eval.evaluate()
eval.accumulate()
eval.summarize()
results.extend(eval.stats[:2]) # update results (mAP@0.5:0.95, mAP@0.5)
map_bbox, map50_bbox, map_mask, map50_mask = results
# Semantic Segmentation
from utils.stuff_seg.cocostuffeval import COCOStuffeval
LOGGER.info(f'\nEvaluating pycocotools stuff... ')
imgIds = [int(x) for x in img_id_list]
stuffGt = COCO(semantic_anno_json) # initialize COCO ground truth api
stuffDt = stuffGt.loadRes(semantic_pred_json) # initialize COCO pred api
cocoStuffEval = COCOStuffeval(stuffGt, stuffDt)
cocoStuffEval.params.imgIds = imgIds # image IDs to evaluate
cocoStuffEval.evaluate()
stats, statsClass = cocoStuffEval.summarize()
stuffIds = getCocoIds(name = 'stuff')
title = ' {:<5} | {:^6} | {:^6} '.format('class', 'iou', 'macc') if (0 >= len(stuff_names)) else \
' {:<5} | {:<20} | {:^6} | {:^6} '.format('class', 'class name', 'iou', 'macc')
print(title)
for idx, (iou, macc) in enumerate(zip(statsClass['ious'], statsClass['maccs'])):
id = (idx + 1)
if id not in stuffIds:
continue
content = ' {:<5} | {:0.4f} | {:0.4f} '.format(str(id), iou, macc) if (0 >= len(stuff_names)) else \
' {:<5} | {:<20} | {:0.4f} | {:0.4f} '.format(str(id), str(stuff_names[getMappingIndex(id, name = 'stuff')]), iou, macc)
print(content)
except Exception as e:
LOGGER.info(f'pycocotools unable to run: {e}')
# Return results
model.float() # for training
if not training:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
final_metric = mp_bbox, mr_bbox, map50_bbox, map_bbox, mp_mask, mr_mask, map50_mask, map_mask, miou_sem, fwiou_sem
return (*final_metric, *(loss.cpu() / len(dataloader)).tolist()), metrics.get_maps(nc), t
def parse_opt():
parser = argparse.ArgumentParser()
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128-pan.yaml', help='dataset.yaml path')
parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolo-pan.pt', help='model path(s)')
parser.add_argument('--batch-size', type=int, default=32, help='batch size')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='inference size (pixels)')
parser.add_argument('--conf-thres', type=float, default=0.001, help='confidence threshold')
parser.add_argument('--iou-thres', type=float, default=0.6, help='NMS IoU threshold')
parser.add_argument('--max-det', type=int, default=300, help='maximum detections per image')
parser.add_argument('--task', default='val', help='train, val, test, speed or study')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
parser.add_argument('--single-cls', action='store_true', help='treat as single-class dataset')
parser.add_argument('--augment', action='store_true', help='augmented inference')
parser.add_argument('--verbose', action='store_true', help='report mAP by class')
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
parser.add_argument('--save-hybrid', action='store_true', help='save label+prediction hybrid results to *.txt')
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
parser.add_argument('--save-json', action='store_true', help='save a COCO-JSON results file')
parser.add_argument('--project', default=ROOT / 'runs/val-pan', help='save results to project/name')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')
opt = parser.parse_args()
opt.data = check_yaml(opt.data) # check YAML
# opt.save_json |= opt.data.endswith('coco.yaml')
opt.save_txt |= opt.save_hybrid
print_args(vars(opt))
return opt
def main(opt):
#check_requirements(requirements=ROOT / 'requirements.txt', exclude=('tensorboard', 'thop'))
if opt.task in ('train', 'val', 'test'): # run normally
if opt.conf_thres > 0.001: # https://github.com/
LOGGER.warning(f'WARNING ⚠️ confidence threshold {opt.conf_thres} > 0.001 produces invalid results')
if opt.save_hybrid:
LOGGER.warning('WARNING ⚠️ --save-hybrid returns high mAP from hybrid labels, not from predictions alone')
run(**vars(opt))
else:
weights = opt.weights if isinstance(opt.weights, list) else [opt.weights]
opt.half = torch.cuda.is_available() and opt.device != 'cpu' # FP16 for fastest results
if opt.task == 'speed': # speed benchmarks
# python val.py --task speed --data coco.yaml --batch 1 --weights yolo.pt...
opt.conf_thres, opt.iou_thres, opt.save_json = 0.25, 0.45, False
for opt.weights in weights:
run(**vars(opt), plots=False)
elif opt.task == 'study': # speed vs mAP benchmarks
# python val.py --task study --data coco.yaml --iou 0.7 --weights yolo.pt...
for opt.weights in weights:
f = f'study_{Path(opt.data).stem}_{Path(opt.weights).stem}.txt' # filename to save to
x, y = list(range(256, 1536 + 128, 128)), [] # x axis (image sizes), y axis
for opt.imgsz in x: # img-size
LOGGER.info(f'\nRunning {f} --imgsz {opt.imgsz}...')
r, _, t = run(**vars(opt), plots=False)
y.append(r + t) # results and times
np.savetxt(f, y, fmt='%10.4g') # save
os.system('zip -r study.zip study_*.txt')
plot_val_study(x=x) # plot
if __name__ == "__main__":
opt = parse_opt()
main(opt)
requirements.txt 0 → 100644
View file @ f969ca34
# requirements
# Usage: pip install -r requirements.txt
# Base ------------------------------------------------------------------------
gitpython
ipython
matplotlib>=3.2.2
numpy>=1.18.5
opencv-python>=4.1.1
Pillow>=7.1.2
psutil
PyYAML>=5.3.1
requests>=2.23.0
scipy>=1.4.1
thop>=0.1.1
torch>=1.7.0
torchvision>=0.8.1
tqdm>=4.64.0
# protobuf<=3.20.1
# Logging ---------------------------------------------------------------------
tensorboard>=2.4.1
# clearml>=1.2.0
# comet
# Plotting --------------------------------------------------------------------
pandas>=1.1.4
seaborn>=0.11.0
# Export ----------------------------------------------------------------------
# coremltools>=6.0
# onnx>=1.9.0
# onnx-simplifier>=0.4.1
# nvidia-pyindex
# nvidia-tensorrt
# scikit-learn<=1.1.2
# tensorflow>=2.4.1
# tensorflowjs>=3.9.0
# openvino-dev
# Deploy ----------------------------------------------------------------------
# tritonclient[all]~=2.24.0
# Extras ----------------------------------------------------------------------
# mss
albumentations>=1.0.3
pycocotools>=2.0
scripts/get_coco.sh 0 → 100644
View file @ f969ca34
#!/bin/bash
# COCO 2017 dataset http://cocodataset.org
# Download command: bash ./scripts/get_coco.sh
# Download/unzip labels
d='./' # unzip directory
url=https://github.com/ultralytics/yolov5/releases/download/v1.0/
f='coco2017labels-segments.zip' # or 'coco2017labels.zip', 68 MB
echo 'Downloading' $url$f ' ...'
curl -L $url$f -o $f && unzip -q $f -d $d && rm $f & # download, unzip, remove in background
# Download/unzip images
d='./coco/images' # unzip directory
url=http://images.cocodataset.org/zips/
f1='train2017.zip' # 19G, 118k images
f2='val2017.zip' # 1G, 5k images
f3='test2017.zip' # 7G, 41k images (optional)
for f in $f1 $f2 $f3; do
echo 'Downloading' $url$f '...'
curl -L $url$f -o $f && unzip -q $f -d $d && rm $f & # download, unzip, remove in background
done
wait # finish background tasks
segment/predict.py 0 → 100644
View file @ f969ca34
import argparse
import os
import platform
import sys
from pathlib import Path
import torch
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
from models.common import DetectMultiBackend
from utils.dataloaders import IMG_FORMATS, VID_FORMATS, LoadImages, LoadScreenshots, LoadStreams
from utils.general import (LOGGER, Profile, check_file, check_img_size, check_imshow, check_requirements, colorstr, cv2,
increment_path, non_max_suppression, print_args, scale_boxes, scale_segments,
strip_optimizer, xyxy2xywh)
from utils.plots import Annotator, colors, save_one_box
from utils.segment.general import masks2segments, process_mask
from utils.torch_utils import select_device, smart_inference_mode
@smart_inference_mode()
def run(
weights=ROOT / 'yolo-seg.pt', # model.pt path(s)
source=ROOT / 'data/images', # file/dir/URL/glob/screen/0(webcam)
data=ROOT / 'data/coco.yaml', # dataset.yaml path
imgsz=(640, 640), # inference size (height, width)
conf_thres=0.25, # confidence threshold
iou_thres=0.45, # NMS IOU threshold
max_det=1000, # maximum detections per image
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
view_img=False, # show results
save_txt=False, # save results to *.txt
save_conf=False, # save confidences in --save-txt labels
save_crop=False, # save cropped prediction boxes
nosave=False, # do not save images/videos
classes=None, # filter by class: --class 0, or --class 0 2 3
agnostic_nms=False, # class-agnostic NMS
augment=False, # augmented inference
visualize=False, # visualize features
update=False, # update all models
project=ROOT / 'runs/predict-seg', # save results to project/name
name='exp', # save results to project/name
exist_ok=False, # existing project/name ok, do not increment
line_thickness=3, # bounding box thickness (pixels)
hide_labels=False, # hide labels
hide_conf=False, # hide confidences
half=False, # use FP16 half-precision inference
dnn=False, # use OpenCV DNN for ONNX inference
vid_stride=1, # video frame-rate stride
retina_masks=False,
):
source = str(source)
save_img = not nosave and not source.endswith('.txt') # save inference images
is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)
is_url = source.lower().startswith(('rtsp://', 'rtmp://', 'http://', 'https://'))
webcam = source.isnumeric() or source.endswith('.txt') or (is_url and not is_file)
screenshot = source.lower().startswith('screen')
if is_url and is_file:
source = check_file(source) # download
# Directories
save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
# Load model
device = select_device(device)
model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
stride, names, pt = model.stride, model.names, model.pt
imgsz = check_img_size(imgsz, s=stride) # check image size
# Dataloader
bs = 1 # batch_size
if webcam:
view_img = check_imshow(warn=True)
dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
bs = len(dataset)
elif screenshot:
dataset = LoadScreenshots(source, img_size=imgsz, stride=stride, auto=pt)
else:
dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
vid_path, vid_writer = [None] * bs, [None] * bs
# Run inference
model.warmup(imgsz=(1 if pt else bs, 3, *imgsz)) # warmup
seen, windows, dt = 0, [], (Profile(), Profile(), Profile())
for path, im, im0s, vid_cap, s in dataset:
with dt[0]:
im = torch.from_numpy(im).to(model.device)
im = im.half() if model.fp16 else im.float() # uint8 to fp16/32
im /= 255 # 0 - 255 to 0.0 - 1.0
if len(im.shape) == 3:
im = im[None] # expand for batch dim
# Inference
with dt[1]:
visualize = increment_path(save_dir / Path(path).stem, mkdir=True) if visualize else False
pred, proto = model(im, augment=augment, visualize=visualize)[:2]
# NMS
with dt[2]:
pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det, nm=32)
# Second-stage classifier (optional)
# pred = utils.general.apply_classifier(pred, classifier_model, im, im0s)
# Process predictions
for i, det in enumerate(pred): # per image
seen += 1
if webcam: # batch_size >= 1
p, im0, frame = path[i], im0s[i].copy(), dataset.count
s += f'{i}: '
else:
p, im0, frame = path, im0s.copy(), getattr(dataset, 'frame', 0)
p = Path(p) # to Path
save_path = str(save_dir / p.name) # im.jpg
txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # im.txt
s += '%gx%g ' % im.shape[2:] # print string
imc = im0.copy() if save_crop else im0 # for save_crop
annotator = Annotator(im0, line_width=line_thickness, example=str(names))
if len(det):
masks = process_mask(proto[i], det[:, 6:], det[:, :4], im.shape[2:], upsample=True) # HWC
det[:, :4] = scale_boxes(im.shape[2:], det[:, :4], im0.shape).round() # rescale boxes to im0 size
# Segments
if save_txt:
segments = reversed(masks2segments(masks))
segments = [scale_segments(im.shape[2:], x, im0.shape, normalize=True) for x in segments]
# Print results
for c in det[:, 5].unique():
n = (det[:, 5] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# Mask plotting
annotator.masks(masks,
colors=[colors(x, True) for x in det[:, 5]],
im_gpu=None if retina_masks else im[i])
# Write results
for j, (*xyxy, conf, cls) in enumerate(reversed(det[:, :6])):
if save_txt: # Write to file
segj = segments[j].reshape(-1) # (n,2) to (n*2)
line = (cls, *segj, conf) if save_conf else (cls, *segj) # label format
with open(f'{txt_path}.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or save_crop or view_img: # Add bbox to image
c = int(cls) # integer class
label = None if hide_labels else (names[c] if hide_conf else f'{names[c]} {conf:.2f}')
annotator.box_label(xyxy, label, color=colors(c, True))
# annotator.draw.polygon(segments[j], outline=colors(c, True), width=3)
if save_crop:
save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True)
# Stream results
im0 = annotator.result()
if view_img:
if platform.system() == 'Linux' and p not in windows:
windows.append(p)
cv2.namedWindow(str(p), cv2.WINDOW_NORMAL | cv2.WINDOW_KEEPRATIO) # allow window resize (Linux)
cv2.resizeWindow(str(p), im0.shape[1], im0.shape[0])
cv2.imshow(str(p), im0)
if cv2.waitKey(1) == ord('q'): # 1 millisecond
exit()
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video' or 'stream'
if vid_path[i] != save_path: # new video
vid_path[i] = save_path
if isinstance(vid_writer[i], cv2.VideoWriter):
vid_writer[i].release() # release previous video writer
if vid_cap: # video
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
else: # stream
fps, w, h = 30, im0.shape[1], im0.shape[0]
save_path = str(Path(save_path).with_suffix('.mp4')) # force *.mp4 suffix on results videos
vid_writer[i] = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))
vid_writer[i].write(im0)
# Print time (inference-only)
LOGGER.info(f"{s}{'' if len(det) else '(no detections), '}{dt[1].dt * 1E3:.1f}ms")
# Print results
t = tuple(x.t / seen * 1E3 for x in dt) # speeds per image
LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}' % t)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
if update:
strip_optimizer(weights[0]) # update model (to fix SourceChangeWarning)
def parse_opt():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolo-seg.pt', help='model path(s)')
parser.add_argument('--source', type=str, default=ROOT / 'data/images', help='file/dir/URL/glob/screen/0(webcam)')
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='(optional) dataset.yaml path')
parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[640], help='inference size h,w')
parser.add_argument('--conf-thres', type=float, default=0.25, help='confidence threshold')
parser.add_argument('--iou-thres', type=float, default=0.45, help='NMS IoU threshold')
parser.add_argument('--max-det', type=int, default=1000, help='maximum detections per image')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--view-img', action='store_true', help='show results')
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes')
parser.add_argument('--nosave', action='store_true', help='do not save images/videos')
parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --classes 0, or --classes 0 2 3')
parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
parser.add_argument('--augment', action='store_true', help='augmented inference')
parser.add_argument('--visualize', action='store_true', help='visualize features')
parser.add_argument('--update', action='store_true', help='update all models')
parser.add_argument('--project', default=ROOT / 'runs/predict-seg', help='save results to project/name')
parser.add_argument('--name', default='exp', help='save results to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--line-thickness', default=3, type=int, help='bounding box thickness (pixels)')
parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels')
parser.add_argument('--hide-conf', default=False, action='store_true', help='hide confidences')
parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')
parser.add_argument('--vid-stride', type=int, default=1, help='video frame-rate stride')
parser.add_argument('--retina-masks', action='store_true', help='whether to plot masks in native resolution')
opt = parser.parse_args()
opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1 # expand
print_args(vars(opt))
return opt
def main(opt):
check_requirements(exclude=('tensorboard', 'thop'))
run(**vars(opt))
if __name__ == "__main__":
opt = parse_opt()
main(opt)
segment/train.py 0 → 100644
View file @ f969ca34
import argparse
import math
import os
import random
import sys
import time
from copy import deepcopy
from datetime import datetime
from pathlib import Path
import numpy as np
import torch
import torch.distributed as dist
import torch.nn as nn
import yaml
from torch.optim import lr_scheduler
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
import segment.val as validate # for end-of-epoch mAP
from models.experimental import attempt_load
from models.yolo import SegmentationModel
from utils.autoanchor import check_anchors
from utils.autobatch import check_train_batch_size
from utils.callbacks import Callbacks
from utils.downloads import attempt_download, is_url
from utils.general import (LOGGER, TQDM_BAR_FORMAT, check_amp, check_dataset, check_file, check_git_info,
check_git_status, check_img_size, check_requirements, check_suffix, check_yaml, colorstr,
get_latest_run, increment_path, init_seeds, intersect_dicts, labels_to_class_weights,
labels_to_image_weights, one_cycle, print_args, print_mutation, strip_optimizer, yaml_save)
from utils.loggers import GenericLogger
from utils.plots import plot_evolve, plot_labels
from utils.segment.dataloaders import create_dataloader
from utils.segment.loss_tal import ComputeLoss
from utils.segment.metrics import KEYS, fitness
from utils.segment.plots import plot_images_and_masks, plot_results_with_masks
from utils.torch_utils import (EarlyStopping, ModelEMA, de_parallel, select_device, smart_DDP, smart_optimizer,
smart_resume, torch_distributed_zero_first)
LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1)) # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
GIT_INFO = None#check_git_info()
def train(hyp, opt, device, callbacks): # hyp is path/to/hyp.yaml or hyp dictionary
save_dir, epochs, batch_size, weights, single_cls, evolve, data, cfg, resume, noval, nosave, workers, freeze, mask_ratio = \
Path(opt.save_dir), opt.epochs, opt.batch_size, opt.weights, opt.single_cls, opt.evolve, opt.data, opt.cfg, \
opt.resume, opt.noval, opt.nosave, opt.workers, opt.freeze, opt.mask_ratio
# callbacks.run('on_pretrain_routine_start')
# Directories
w = save_dir / 'weights' # weights dir
(w.parent if evolve else w).mkdir(parents=True, exist_ok=True) # make dir
last, best = w / 'last.pt', w / 'best.pt'
# Hyperparameters
if isinstance(hyp, str):
with open(hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
LOGGER.info(colorstr('hyperparameters: ') + ', '.join(f'{k}={v}' for k, v in hyp.items()))
opt.hyp = hyp.copy() # for saving hyps to checkpoints
# Save run settings
if not evolve:
yaml_save(save_dir / 'hyp.yaml', hyp)
yaml_save(save_dir / 'opt.yaml', vars(opt))
# Loggers
data_dict = None
if RANK in {-1, 0}:
logger = GenericLogger(opt=opt, console_logger=LOGGER)
# Config
plots = not evolve and not opt.noplots # create plots
overlap = not opt.no_overlap
cuda = device.type != 'cpu'
init_seeds(opt.seed + 1 + RANK, deterministic=True)
with torch_distributed_zero_first(LOCAL_RANK):
data_dict = data_dict or check_dataset(data) # check if None
train_path, val_path = data_dict['train'], data_dict['val']
nc = 1 if single_cls else int(data_dict['nc']) # number of classes
names = {0: 'item'} if single_cls and len(data_dict['names']) != 1 else data_dict['names'] # class names
#is_coco = isinstance(val_path, str) and val_path.endswith('coco/val2017.txt') # COCO dataset
is_coco = isinstance(val_path, str) and val_path.endswith('val2017.txt') # COCO dataset
# Model
check_suffix(weights, '.pt') # check weights
pretrained = weights.endswith('.pt')
if pretrained:
with torch_distributed_zero_first(LOCAL_RANK):
weights = attempt_download(weights) # download if not found locally
ckpt = torch.load(weights, map_location='cpu') # load checkpoint to CPU to avoid CUDA memory leak
model = SegmentationModel(cfg or ckpt['model'].yaml, ch=3, nc=nc).to(device)
exclude = ['anchor'] if (cfg or hyp.get('anchors')) and not resume else [] # exclude keys
csd = ckpt['model'].float().state_dict() # checkpoint state_dict as FP32
csd = intersect_dicts(csd, model.state_dict(), exclude=exclude) # intersect
model.load_state_dict(csd, strict=False) # load
LOGGER.info(f'Transferred {len(csd)}/{len(model.state_dict())} items from {weights}') # report
else:
model = SegmentationModel(cfg, ch=3, nc=nc).to(device) # create
amp = check_amp(model) # check AMP
# Freeze
freeze = [f'model.{x}.' for x in (freeze if len(freeze) > 1 else range(freeze[0]))] # layers to freeze
for k, v in model.named_parameters():
#v.requires_grad = True # train all layers
# v.register_hook(lambda x: torch.nan_to_num(x)) # NaN to 0 (commented for erratic training results)
if any(x in k for x in freeze):
LOGGER.info(f'freezing {k}')
v.requires_grad = False
# Image size
gs = max(int(model.stride.max()), 32) # grid size (max stride)
imgsz = check_img_size(opt.imgsz, gs, floor=gs * 2) # verify imgsz is gs-multiple
# Batch size
if RANK == -1 and batch_size == -1: # single-GPU only, estimate best batch size
batch_size = check_train_batch_size(model, imgsz, amp)
logger.update_params({"batch_size": batch_size})
# loggers.on_params_update({"batch_size": batch_size})
# Optimizer
nbs = 64 # nominal batch size
accumulate = max(round(nbs / batch_size), 1) # accumulate loss before optimizing
hyp['weight_decay'] *= batch_size * accumulate / nbs # scale weight_decay
optimizer = smart_optimizer(model, opt.optimizer, hyp['lr0'], hyp['momentum'], hyp['weight_decay'])
# Scheduler
if opt.cos_lr:
lf = one_cycle(1, hyp['lrf'], epochs) # cosine 1->hyp['lrf']
else:
lf = lambda x: (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf'] # linear
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf) # plot_lr_scheduler(optimizer, scheduler, epochs)
# EMA
ema = ModelEMA(model) if RANK in {-1, 0} else None
# Resume
best_fitness, start_epoch = 0.0, 0
if pretrained:
if resume:
best_fitness, start_epoch, epochs = smart_resume(ckpt, optimizer, ema, weights, epochs, resume)
del ckpt, csd
# DP mode
if cuda and RANK == -1 and torch.cuda.device_count() > 1:
LOGGER.warning('WARNING ⚠️ DP not recommended, use torch.distributed.run for best DDP Multi-GPU results.')
model = torch.nn.DataParallel(model)
# SyncBatchNorm
if opt.sync_bn and cuda and RANK != -1:
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
LOGGER.info('Using SyncBatchNorm()')
# Trainloader
train_loader, dataset = create_dataloader(
train_path,
imgsz,
batch_size // WORLD_SIZE,
gs,
single_cls,
hyp=hyp,
augment=True,
cache=None if opt.cache == 'val' else opt.cache,
rect=opt.rect,
rank=LOCAL_RANK,
workers=workers,
image_weights=opt.image_weights,
close_mosaic=opt.close_mosaic != 0,
quad=opt.quad,
prefix=colorstr('train: '),
shuffle=True,
mask_downsample_ratio=mask_ratio,
overlap_mask=overlap,
)
labels = np.concatenate(dataset.labels, 0)
mlc = int(labels[:, 0].max()) # max label class
assert mlc < nc, f'Label class {mlc} exceeds nc={nc} in {data}. Possible class labels are 0-{nc - 1}'
# Process 0
if RANK in {-1, 0}:
val_loader = create_dataloader(val_path,
imgsz,
batch_size // WORLD_SIZE * 2,
gs,
single_cls,
hyp=hyp,
cache=None if noval else opt.cache,
rect=True,
rank=-1,
workers=workers * 2,
pad=0.5,
mask_downsample_ratio=mask_ratio,
overlap_mask=overlap,
prefix=colorstr('val: '))[0]
if not resume:
#if not opt.noautoanchor:
# check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz) # run AutoAnchor
model.half().float() # pre-reduce anchor precision
if plots:
plot_labels(labels, names, save_dir)
# callbacks.run('on_pretrain_routine_end', labels, names)
# DDP mode
if cuda and RANK != -1:
model = smart_DDP(model)
# Model attributes
nl = de_parallel(model).model[-1].nl # number of detection layers (to scale hyps)
#hyp['box'] *= 3 / nl # scale to layers
#hyp['cls'] *= nc / 80 * 3 / nl # scale to classes and layers
#hyp['obj'] *= (imgsz / 640) ** 2 * 3 / nl # scale to image size and layers
hyp['label_smoothing'] = opt.label_smoothing
model.nc = nc # attach number of classes to model
model.hyp = hyp # attach hyperparameters to model
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) * nc # attach class weights
model.names = names
# Start training
t0 = time.time()
nb = len(train_loader) # number of batches
nw = max(round(hyp['warmup_epochs'] * nb), 100) # number of warmup iterations, max(3 epochs, 100 iterations)
# nw = min(nw, (epochs - start_epoch) / 2 * nb) # limit warmup to < 1/2 of training
last_opt_step = -1
maps = np.zeros(nc) # mAP per class
results = (0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
scheduler.last_epoch = start_epoch - 1 # do not move
scaler = torch.cuda.amp.GradScaler(enabled=amp)
stopper, stop = EarlyStopping(patience=opt.patience), False
compute_loss = ComputeLoss(model, overlap=overlap) # init loss class
# callbacks.run('on_train_start')
LOGGER.info(f'Image sizes {imgsz} train, {imgsz} val\n'
f'Using {train_loader.num_workers * WORLD_SIZE} dataloader workers\n'
f"Logging results to {colorstr('bold', save_dir)}\n"
f'Starting training for {epochs} epochs...')
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
# callbacks.run('on_train_epoch_start')
model.train()
# Update image weights (optional, single-GPU only)
if opt.image_weights:
cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 / nc # class weights
iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw) # image weights
dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n) # rand weighted idx
if epoch == (epochs - opt.close_mosaic):
LOGGER.info("Closing dataloader mosaic")
dataset.mosaic = False
# Update mosaic border (optional)
# b = int(random.uniform(0.25 * imgsz, 0.75 * imgsz + gs) // gs * gs)
# dataset.mosaic_border = [b - imgsz, -b] # height, width borders
mloss = torch.zeros(4, device=device) # mean losses
if RANK != -1:
train_loader.sampler.set_epoch(epoch)
pbar = enumerate(train_loader)
LOGGER.info(('\n' + '%11s' * 8) %
('Epoch', 'GPU_mem', 'box_loss', 'seg_loss', 'cls_loss', 'dfl_loss', 'Instances', 'Size'))
if RANK in {-1, 0}:
pbar = tqdm(pbar, total=nb, bar_format=TQDM_BAR_FORMAT) # progress bar
optimizer.zero_grad()
for i, (imgs, targets, paths, _, masks) in pbar: # batch ------------------------------------------------------
# callbacks.run('on_train_batch_start')
ni = i + nb * epoch # number integrated batches (since train start)
imgs = imgs.to(device, non_blocking=True).float() / 255 # uint8 to float32, 0-255 to 0.0-1.0
# Warmup
if ni <= nw:
xi = [0, nw] # x interp
# compute_loss.gr = np.interp(ni, xi, [0.0, 1.0]) # iou loss ratio (obj_loss = 1.0 or iou)
accumulate = max(1, np.interp(ni, xi, [1, nbs / batch_size]).round())
for j, x in enumerate(optimizer.param_groups):
# bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
x['lr'] = np.interp(ni, xi, [hyp['warmup_bias_lr'] if j == 0 else 0.0, x['initial_lr'] * lf(epoch)])
if 'momentum' in x:
x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])
# Multi-scale
if opt.multi_scale:
sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs # size
sf = sz / max(imgs.shape[2:]) # scale factor
if sf != 1:
ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]] # new shape (stretched to gs-multiple)
imgs = nn.functional.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)
# Forward
with torch.cuda.amp.autocast(amp):
pred = model(imgs) # forward
loss, loss_items = compute_loss(pred, targets.to(device), masks=masks.to(device).float())
if RANK != -1:
loss *= WORLD_SIZE # gradient averaged between devices in DDP mode
if opt.quad:
loss *= 4.
# Backward
scaler.scale(loss).backward()
# Optimize - https://pytorch.org/docs/master/notes/amp_examples.html
if ni - last_opt_step >= accumulate:
scaler.unscale_(optimizer) # unscale gradients
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=10.0) # clip gradients
scaler.step(optimizer) # optimizer.step
scaler.update()
optimizer.zero_grad()
if ema:
ema.update(model)
last_opt_step = ni
# Log
if RANK in {-1, 0}:
mloss = (mloss * i + loss_items) / (i + 1) # update mean losses
mem = f'{torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0:.3g}G' # (GB)
pbar.set_description(('%11s' * 2 + '%11.4g' * 6) %
(f'{epoch}/{epochs - 1}', mem, *mloss, targets.shape[0], imgs.shape[-1]))
# callbacks.run('on_train_batch_end', model, ni, imgs, targets, paths)
# if callbacks.stop_training:
# return
# Mosaic plots
if plots:
if ni < 3:
plot_images_and_masks(imgs, targets, masks, paths, save_dir / f"train_batch{ni}.jpg")
if ni == 10:
files = sorted(save_dir.glob('train*.jpg'))
logger.log_images(files, "Mosaics", epoch)
# end batch ------------------------------------------------------------------------------------------------
# Scheduler
lr = [x['lr'] for x in optimizer.param_groups] # for loggers
scheduler.step()
if RANK in {-1, 0}:
# mAP
# callbacks.run('on_train_epoch_end', epoch=epoch)
ema.update_attr(model, include=['yaml', 'nc', 'hyp', 'names', 'stride', 'class_weights'])
final_epoch = (epoch + 1 == epochs) or stopper.possible_stop
if not noval or final_epoch: # Calculate mAP
results, maps, _ = validate.run(data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
half=amp,
model=ema.ema,
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
plots=False,
callbacks=callbacks,
compute_loss=compute_loss,
mask_downsample_ratio=mask_ratio,
overlap=overlap)
# Update best mAP
fi = fitness(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
stop = stopper(epoch=epoch, fitness=fi) # early stop check
if fi > best_fitness:
best_fitness = fi
log_vals = list(mloss) + list(results) + lr
# callbacks.run('on_fit_epoch_end', log_vals, epoch, best_fitness, fi)
# Log val metrics and media
metrics_dict = dict(zip(KEYS, log_vals))
logger.log_metrics(metrics_dict, epoch)
# Save model
if (not nosave) or (final_epoch and not evolve): # if save
ckpt = {
'epoch': epoch,
'best_fitness': best_fitness,
'model': deepcopy(de_parallel(model)).half(),
'ema': deepcopy(ema.ema).half(),
'updates': ema.updates,
'optimizer': optimizer.state_dict(),
'opt': vars(opt),
'git': GIT_INFO, # {remote, branch, commit} if a git repo
'date': datetime.now().isoformat()}
# Save last, best and delete
torch.save(ckpt, last)
if best_fitness == fi:
torch.save(ckpt, best)
if opt.save_period > 0 and epoch % opt.save_period == 0:
torch.save(ckpt, w / f'epoch{epoch}.pt')
logger.log_model(w / f'epoch{epoch}.pt')
del ckpt
# callbacks.run('on_model_save', last, epoch, final_epoch, best_fitness, fi)
# EarlyStopping
if RANK != -1: # if DDP training
broadcast_list = [stop if RANK == 0 else None]
dist.broadcast_object_list(broadcast_list, 0) # broadcast 'stop' to all ranks
if RANK != 0:
stop = broadcast_list[0]
if stop:
break # must break all DDP ranks
# end epoch ----------------------------------------------------------------------------------------------------
# end training -----------------------------------------------------------------------------------------------------
if RANK in {-1, 0}:
LOGGER.info(f'\n{epoch - start_epoch + 1} epochs completed in {(time.time() - t0) / 3600:.3f} hours.')
for f in last, best:
if f.exists():
strip_optimizer(f) # strip optimizers
if f is best:
LOGGER.info(f'\nValidating {f}...')
results, _, _ = validate.run(
data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
model=attempt_load(f, device).half(),
iou_thres=0.65 if is_coco else 0.60, # best pycocotools at iou 0.65
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
save_json=is_coco,
verbose=True,
plots=plots,
callbacks=callbacks,
compute_loss=compute_loss,
mask_downsample_ratio=mask_ratio,
overlap=overlap) # val best model with plots
if is_coco:
# callbacks.run('on_fit_epoch_end', list(mloss) + list(results) + lr, epoch, best_fitness, fi)
metrics_dict = dict(zip(KEYS, list(mloss) + list(results) + lr))
logger.log_metrics(metrics_dict, epoch)
# callbacks.run('on_train_end', last, best, epoch, results)
# on train end callback using genericLogger
logger.log_metrics(dict(zip(KEYS[4:16], results)), epochs)
if not opt.evolve:
logger.log_model(best, epoch)
if plots:
plot_results_with_masks(file=save_dir / 'results.csv') # save results.png
files = ['results.png', 'confusion_matrix.png', *(f'{x}_curve.png' for x in ('F1', 'PR', 'P', 'R'))]
files = [(save_dir / f) for f in files if (save_dir / f).exists()] # filter
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}")
logger.log_images(files, "Results", epoch + 1)
logger.log_images(sorted(save_dir.glob('val*.jpg')), "Validation", epoch + 1)
torch.cuda.empty_cache()
return results
def parse_opt(known=False):
parser = argparse.ArgumentParser()
parser.add_argument('--weights', type=str, default=ROOT / 'yolo-seg.pt', help='initial weights path')
parser.add_argument('--cfg', type=str, default='', help='model.yaml path')
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128-seg.yaml', help='dataset.yaml path')
parser.add_argument('--hyp', type=str, default=ROOT / 'data/hyps/hyp.scratch-low.yaml', help='hyperparameters path')
parser.add_argument('--epochs', type=int, default=100, help='total training epochs')
parser.add_argument('--batch-size', type=int, default=16, help='total batch size for all GPUs, -1 for autobatch')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='train, val image size (pixels)')
parser.add_argument('--rect', action='store_true', help='rectangular training')
parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
parser.add_argument('--noval', action='store_true', help='only validate final epoch')
parser.add_argument('--noautoanchor', action='store_true', help='disable AutoAnchor')
parser.add_argument('--noplots', action='store_true', help='save no plot files')
parser.add_argument('--evolve', type=int, nargs='?', const=300, help='evolve hyperparameters for x generations')
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
parser.add_argument('--cache', type=str, nargs='?', const='ram', help='image --cache ram/disk')
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
parser.add_argument('--single-cls', action='store_true', help='train multi-class data as single-class')
parser.add_argument('--optimizer', type=str, choices=['SGD', 'Adam', 'AdamW', 'LION'], default='SGD', help='optimizer')
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
parser.add_argument('--project', default=ROOT / 'runs/train-seg', help='save to project/name')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--quad', action='store_true', help='quad dataloader')
parser.add_argument('--cos-lr', action='store_true', help='cosine LR scheduler')
parser.add_argument('--label-smoothing', type=float, default=0.0, help='Label smoothing epsilon')
parser.add_argument('--patience', type=int, default=100, help='EarlyStopping patience (epochs without improvement)')
parser.add_argument('--freeze', nargs='+', type=int, default=[0], help='Freeze layers: backbone=10, first3=0 1 2')
parser.add_argument('--save-period', type=int, default=-1, help='Save checkpoint every x epochs (disabled if < 1)')
parser.add_argument('--seed', type=int, default=0, help='Global training seed')
parser.add_argument('--local_rank', type=int, default=-1, help='Automatic DDP Multi-GPU argument, do not modify')
parser.add_argument('--close-mosaic', type=int, default=0, help='Experimental')
# Instance Segmentation Args
parser.add_argument('--mask-ratio', type=int, default=4, help='Downsample the truth masks to saving memory')
parser.add_argument('--no-overlap', action='store_true', help='Overlap masks train faster at slightly less mAP')
return parser.parse_known_args()[0] if known else parser.parse_args()
def main(opt, callbacks=Callbacks()):
# Checks
if RANK in {-1, 0}:
print_args(vars(opt))
#check_git_status()
#check_requirements()
# Resume
if opt.resume and not opt.evolve: # resume from specified or most recent last.pt
last = Path(check_file(opt.resume) if isinstance(opt.resume, str) else get_latest_run())
opt_yaml = last.parent.parent / 'opt.yaml' # train options yaml
opt_data = opt.data # original dataset
if opt_yaml.is_file():
with open(opt_yaml, errors='ignore') as f:
d = yaml.safe_load(f)
else:
d = torch.load(last, map_location='cpu')['opt']
opt = argparse.Namespace(**d) # replace
opt.cfg, opt.weights, opt.resume = '', str(last), True # reinstate
if is_url(opt_data):
opt.data = check_file(opt_data) # avoid HUB resume auth timeout
else:
opt.data, opt.cfg, opt.hyp, opt.weights, opt.project = \
check_file(opt.data), check_yaml(opt.cfg), check_yaml(opt.hyp), str(opt.weights), str(opt.project) # checks
assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
if opt.evolve:
if opt.project == str(ROOT / 'runs/train'): # if default project name, rename to runs/evolve
opt.project = str(ROOT / 'runs/evolve')
opt.exist_ok, opt.resume = opt.resume, False # pass resume to exist_ok and disable resume
if opt.name == 'cfg':
opt.name = Path(opt.cfg).stem # use model.yaml as name
opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok))
# DDP mode
device = select_device(opt.device, batch_size=opt.batch_size)
if LOCAL_RANK != -1:
msg = 'is not compatible with YOLO Multi-GPU DDP training'
assert not opt.image_weights, f'--image-weights {msg}'
assert not opt.evolve, f'--evolve {msg}'
assert opt.batch_size != -1, f'AutoBatch with --batch-size -1 {msg}, please pass a valid --batch-size'
assert opt.batch_size % WORLD_SIZE == 0, f'--batch-size {opt.batch_size} must be multiple of WORLD_SIZE'
assert torch.cuda.device_count() > LOCAL_RANK, 'insufficient CUDA devices for DDP command'
torch.cuda.set_device(LOCAL_RANK)
device = torch.device('cuda', LOCAL_RANK)
dist.init_process_group(backend="nccl" if dist.is_nccl_available() else "gloo")
# Train
if not opt.evolve:
train(opt.hyp, opt, device, callbacks)
# Evolve hyperparameters (optional)
else:
# Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
meta = {
'lr0': (1, 1e-5, 1e-1), # initial learning rate (SGD=1E-2, Adam=1E-3)
'lrf': (1, 0.01, 1.0), # final OneCycleLR learning rate (lr0 * lrf)
'momentum': (0.3, 0.6, 0.98), # SGD momentum/Adam beta1
'weight_decay': (1, 0.0, 0.001), # optimizer weight decay
'warmup_epochs': (1, 0.0, 5.0), # warmup epochs (fractions ok)
'warmup_momentum': (1, 0.0, 0.95), # warmup initial momentum
'warmup_bias_lr': (1, 0.0, 0.2), # warmup initial bias lr
'box': (1, 0.02, 0.2), # box loss gain
'cls': (1, 0.2, 4.0), # cls loss gain
'cls_pw': (1, 0.5, 2.0), # cls BCELoss positive_weight
'obj': (1, 0.2, 4.0), # obj loss gain (scale with pixels)
'obj_pw': (1, 0.5, 2.0), # obj BCELoss positive_weight
'iou_t': (0, 0.1, 0.7), # IoU training threshold
'anchor_t': (1, 2.0, 8.0), # anchor-multiple threshold
'anchors': (2, 2.0, 10.0), # anchors per output grid (0 to ignore)
'fl_gamma': (0, 0.0, 2.0), # focal loss gamma (efficientDet default gamma=1.5)
'hsv_h': (1, 0.0, 0.1), # image HSV-Hue augmentation (fraction)
'hsv_s': (1, 0.0, 0.9), # image HSV-Saturation augmentation (fraction)
'hsv_v': (1, 0.0, 0.9), # image HSV-Value augmentation (fraction)
'degrees': (1, 0.0, 45.0), # image rotation (+/- deg)
'translate': (1, 0.0, 0.9), # image translation (+/- fraction)
'scale': (1, 0.0, 0.9), # image scale (+/- gain)
'shear': (1, 0.0, 10.0), # image shear (+/- deg)
'perspective': (0, 0.0, 0.001), # image perspective (+/- fraction), range 0-0.001
'flipud': (1, 0.0, 1.0), # image flip up-down (probability)
'fliplr': (0, 0.0, 1.0), # image flip left-right (probability)
'mosaic': (1, 0.0, 1.0), # image mixup (probability)
'mixup': (1, 0.0, 1.0), # image mixup (probability)
'copy_paste': (1, 0.0, 1.0)} # segment copy-paste (probability)
with open(opt.hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
if 'anchors' not in hyp: # anchors commented in hyp.yaml
hyp['anchors'] = 3
if opt.noautoanchor:
del hyp['anchors'], meta['anchors']
opt.noval, opt.nosave, save_dir = True, True, Path(opt.save_dir) # only val/save final epoch
# ei = [isinstance(x, (int, float)) for x in hyp.values()] # evolvable indices
evolve_yaml, evolve_csv = save_dir / 'hyp_evolve.yaml', save_dir / 'evolve.csv'
if opt.bucket:
os.system(f'gsutil cp gs://{opt.bucket}/evolve.csv {evolve_csv}') # download evolve.csv if exists
for _ in range(opt.evolve): # generations to evolve
if evolve_csv.exists(): # if evolve.csv exists: select best hyps and mutate
# Select parent(s)
parent = 'single' # parent selection method: 'single' or 'weighted'
x = np.loadtxt(evolve_csv, ndmin=2, delimiter=',', skiprows=1)
n = min(5, len(x)) # number of previous results to consider
x = x[np.argsort(-fitness(x))][:n] # top n mutations
w = fitness(x) - fitness(x).min() + 1E-6 # weights (sum > 0)
if parent == 'single' or len(x) == 1:
# x = x[random.randint(0, n - 1)] # random selection
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
elif parent == 'weighted':
x = (x * w.reshape(n, 1)).sum(0) / w.sum() # weighted combination
# Mutate
mp, s = 0.8, 0.2 # mutation probability, sigma
npr = np.random
npr.seed(int(time.time()))
g = np.array([meta[k][0] for k in hyp.keys()]) # gains 0-1
ng = len(meta)
v = np.ones(ng)
while all(v == 1): # mutate until a change occurs (prevent duplicates)
v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
for i, k in enumerate(hyp.keys()): # plt.hist(v.ravel(), 300)
hyp[k] = float(x[i + 7] * v[i]) # mutate
# Constrain to limits
for k, v in meta.items():
hyp[k] = max(hyp[k], v[1]) # lower limit
hyp[k] = min(hyp[k], v[2]) # upper limit
hyp[k] = round(hyp[k], 5) # significant digits
# Train mutation
results = train(hyp.copy(), opt, device, callbacks)
callbacks = Callbacks()
# Write mutation results
print_mutation(KEYS, results, hyp.copy(), save_dir, opt.bucket)
# Plot results
plot_evolve(evolve_csv)
LOGGER.info(f'Hyperparameter evolution finished {opt.evolve} generations\n'
f"Results saved to {colorstr('bold', save_dir)}\n"
f'Usage example: $ python train.py --hyp {evolve_yaml}')
def run(**kwargs):
# Usage: import train; train.run(data='coco128.yaml', imgsz=320, weights='yolo.pt')
opt = parse_opt(True)
for k, v in kwargs.items():
setattr(opt, k, v)
main(opt)
return opt
if __name__ == "__main__":
opt = parse_opt()
main(opt)
segment/train_dual.py 0 → 100644
View file @ f969ca34
import argparse
import math
import os
import random
import sys
import time
from copy import deepcopy
from datetime import datetime
from pathlib import Path
import numpy as np
import torch
import torch.distributed as dist
import torch.nn as nn
import yaml
from torch.optim import lr_scheduler
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
import segment.val_dual as validate # for end-of-epoch mAP
from models.experimental import attempt_load
from models.yolo import SegmentationModel
from utils.autoanchor import check_anchors
from utils.autobatch import check_train_batch_size
from utils.callbacks import Callbacks
from utils.downloads import attempt_download, is_url
from utils.general import (LOGGER, TQDM_BAR_FORMAT, check_amp, check_dataset, check_file, check_git_info,
check_git_status, check_img_size, check_requirements, check_suffix, check_yaml, colorstr,
get_latest_run, increment_path, init_seeds, intersect_dicts, labels_to_class_weights,
labels_to_image_weights, one_cycle, print_args, print_mutation, strip_optimizer, yaml_save)
from utils.loggers import GenericLogger
from utils.plots import plot_evolve, plot_labels
from utils.segment.dataloaders import create_dataloader
from utils.segment.loss_tal_dual import ComputeLoss
#from utils.segment.loss_tal_dual import ComputeLossLH as ComputeLoss
from utils.segment.metrics import KEYS, fitness
from utils.segment.plots import plot_images_and_masks, plot_results_with_masks
from utils.torch_utils import (EarlyStopping, ModelEMA, de_parallel, select_device, smart_DDP, smart_optimizer,
smart_resume, torch_distributed_zero_first)
LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1)) # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
GIT_INFO = None#check_git_info()
def train(hyp, opt, device, callbacks): # hyp is path/to/hyp.yaml or hyp dictionary
save_dir, epochs, batch_size, weights, single_cls, evolve, data, cfg, resume, noval, nosave, workers, freeze, mask_ratio = \
Path(opt.save_dir), opt.epochs, opt.batch_size, opt.weights, opt.single_cls, opt.evolve, opt.data, opt.cfg, \
opt.resume, opt.noval, opt.nosave, opt.workers, opt.freeze, opt.mask_ratio
# callbacks.run('on_pretrain_routine_start')
# Directories
w = save_dir / 'weights' # weights dir
(w.parent if evolve else w).mkdir(parents=True, exist_ok=True) # make dir
last, best = w / 'last.pt', w / 'best.pt'
# Hyperparameters
if isinstance(hyp, str):
with open(hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
LOGGER.info(colorstr('hyperparameters: ') + ', '.join(f'{k}={v}' for k, v in hyp.items()))
opt.hyp = hyp.copy() # for saving hyps to checkpoints
# Save run settings
if not evolve:
yaml_save(save_dir / 'hyp.yaml', hyp)
yaml_save(save_dir / 'opt.yaml', vars(opt))
# Loggers
data_dict = None
if RANK in {-1, 0}:
logger = GenericLogger(opt=opt, console_logger=LOGGER)
# Config
plots = not evolve and not opt.noplots # create plots
overlap = not opt.no_overlap
cuda = device.type != 'cpu'
init_seeds(opt.seed + 1 + RANK, deterministic=True)
with torch_distributed_zero_first(LOCAL_RANK):
data_dict = data_dict or check_dataset(data) # check if None
train_path, val_path = data_dict['train'], data_dict['val']
nc = 1 if single_cls else int(data_dict['nc']) # number of classes
names = {0: 'item'} if single_cls and len(data_dict['names']) != 1 else data_dict['names'] # class names
#is_coco = isinstance(val_path, str) and val_path.endswith('coco/val2017.txt') # COCO dataset
is_coco = isinstance(val_path, str) and val_path.endswith('val2017.txt') # COCO dataset
# Model
check_suffix(weights, '.pt') # check weights
pretrained = weights.endswith('.pt')
if pretrained:
with torch_distributed_zero_first(LOCAL_RANK):
weights = attempt_download(weights) # download if not found locally
ckpt = torch.load(weights, map_location='cpu') # load checkpoint to CPU to avoid CUDA memory leak
model = SegmentationModel(cfg or ckpt['model'].yaml, ch=3, nc=nc).to(device)
exclude = ['anchor'] if (cfg or hyp.get('anchors')) and not resume else [] # exclude keys
csd = ckpt['model'].float().state_dict() # checkpoint state_dict as FP32
csd = intersect_dicts(csd, model.state_dict(), exclude=exclude) # intersect
model.load_state_dict(csd, strict=False) # load
LOGGER.info(f'Transferred {len(csd)}/{len(model.state_dict())} items from {weights}') # report
else:
model = SegmentationModel(cfg, ch=3, nc=nc).to(device) # create
amp = check_amp(model) # check AMP
# Freeze
freeze = [f'model.{x}.' for x in (freeze if len(freeze) > 1 else range(freeze[0]))] # layers to freeze
for k, v in model.named_parameters():
#v.requires_grad = True # train all layers
# v.register_hook(lambda x: torch.nan_to_num(x)) # NaN to 0 (commented for erratic training results)
if any(x in k for x in freeze):
LOGGER.info(f'freezing {k}')
v.requires_grad = False
# Image size
gs = max(int(model.stride.max()), 32) # grid size (max stride)
imgsz = check_img_size(opt.imgsz, gs, floor=gs * 2) # verify imgsz is gs-multiple
# Batch size
if RANK == -1 and batch_size == -1: # single-GPU only, estimate best batch size
batch_size = check_train_batch_size(model, imgsz, amp)
logger.update_params({"batch_size": batch_size})
# loggers.on_params_update({"batch_size": batch_size})
# Optimizer
nbs = 64 # nominal batch size
accumulate = max(round(nbs / batch_size), 1) # accumulate loss before optimizing
hyp['weight_decay'] *= batch_size * accumulate / nbs # scale weight_decay
optimizer = smart_optimizer(model, opt.optimizer, hyp['lr0'], hyp['momentum'], hyp['weight_decay'])
# Scheduler
if opt.cos_lr:
lf = one_cycle(1, hyp['lrf'], epochs) # cosine 1->hyp['lrf']
else:
lf = lambda x: (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf'] # linear
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf) # plot_lr_scheduler(optimizer, scheduler, epochs)
# EMA
ema = ModelEMA(model) if RANK in {-1, 0} else None
# Resume
best_fitness, start_epoch = 0.0, 0
if pretrained:
if resume:
best_fitness, start_epoch, epochs = smart_resume(ckpt, optimizer, ema, weights, epochs, resume)
del ckpt, csd
# DP mode
if cuda and RANK == -1 and torch.cuda.device_count() > 1:
LOGGER.warning('WARNING ⚠️ DP not recommended, use torch.distributed.run for best DDP Multi-GPU results.')
model = torch.nn.DataParallel(model)
# SyncBatchNorm
if opt.sync_bn and cuda and RANK != -1:
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
LOGGER.info('Using SyncBatchNorm()')
# Trainloader
train_loader, dataset = create_dataloader(
train_path,
imgsz,
batch_size // WORLD_SIZE,
gs,
single_cls,
hyp=hyp,
augment=True,
cache=None if opt.cache == 'val' else opt.cache,
rect=opt.rect,
rank=LOCAL_RANK,
workers=workers,
image_weights=opt.image_weights,
close_mosaic=opt.close_mosaic != 0,
quad=opt.quad,
prefix=colorstr('train: '),
shuffle=True,
mask_downsample_ratio=mask_ratio,
overlap_mask=overlap,
)
labels = np.concatenate(dataset.labels, 0)
mlc = int(labels[:, 0].max()) # max label class
assert mlc < nc, f'Label class {mlc} exceeds nc={nc} in {data}. Possible class labels are 0-{nc - 1}'
# Process 0
if RANK in {-1, 0}:
val_loader = create_dataloader(val_path,
imgsz,
batch_size // WORLD_SIZE * 2,
gs,
single_cls,
hyp=hyp,
cache=None if noval else opt.cache,
rect=True,
rank=-1,
workers=workers * 2,
pad=0.5,
mask_downsample_ratio=mask_ratio,
overlap_mask=overlap,
prefix=colorstr('val: '))[0]
if not resume:
#if not opt.noautoanchor:
# check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz) # run AutoAnchor
model.half().float() # pre-reduce anchor precision
if plots:
plot_labels(labels, names, save_dir)
# callbacks.run('on_pretrain_routine_end', labels, names)
# DDP mode
if cuda and RANK != -1:
model = smart_DDP(model)
# Model attributes
nl = de_parallel(model).model[-1].nl # number of detection layers (to scale hyps)
#hyp['box'] *= 3 / nl # scale to layers
#hyp['cls'] *= nc / 80 * 3 / nl # scale to classes and layers
#hyp['obj'] *= (imgsz / 640) ** 2 * 3 / nl # scale to image size and layers
hyp['label_smoothing'] = opt.label_smoothing
model.nc = nc # attach number of classes to model
model.hyp = hyp # attach hyperparameters to model
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) * nc # attach class weights
model.names = names
# Start training
t0 = time.time()
nb = len(train_loader) # number of batches
nw = max(round(hyp['warmup_epochs'] * nb), 100) # number of warmup iterations, max(3 epochs, 100 iterations)
# nw = min(nw, (epochs - start_epoch) / 2 * nb) # limit warmup to < 1/2 of training
last_opt_step = -1
maps = np.zeros(nc) # mAP per class
results = (0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
scheduler.last_epoch = start_epoch - 1 # do not move
scaler = torch.cuda.amp.GradScaler(enabled=amp)
stopper, stop = EarlyStopping(patience=opt.patience), False
compute_loss = ComputeLoss(model, overlap=overlap) # init loss class
# callbacks.run('on_train_start')
LOGGER.info(f'Image sizes {imgsz} train, {imgsz} val\n'
f'Using {train_loader.num_workers * WORLD_SIZE} dataloader workers\n'
f"Logging results to {colorstr('bold', save_dir)}\n"
f'Starting training for {epochs} epochs...')
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
# callbacks.run('on_train_epoch_start')
model.train()
# Update image weights (optional, single-GPU only)
if opt.image_weights:
cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 / nc # class weights
iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw) # image weights
dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n) # rand weighted idx
if epoch == (epochs - opt.close_mosaic):
LOGGER.info("Closing dataloader mosaic")
dataset.mosaic = False
# Update mosaic border (optional)
# b = int(random.uniform(0.25 * imgsz, 0.75 * imgsz + gs) // gs * gs)
# dataset.mosaic_border = [b - imgsz, -b] # height, width borders
mloss = torch.zeros(4, device=device) # mean losses
if RANK != -1:
train_loader.sampler.set_epoch(epoch)
pbar = enumerate(train_loader)
LOGGER.info(('\n' + '%11s' * 8) %
('Epoch', 'GPU_mem', 'box_loss', 'seg_loss', 'cls_loss', 'dfl_loss', 'Instances', 'Size'))
if RANK in {-1, 0}:
pbar = tqdm(pbar, total=nb, bar_format=TQDM_BAR_FORMAT) # progress bar
optimizer.zero_grad()
for i, (imgs, targets, paths, _, masks) in pbar: # batch ------------------------------------------------------
# callbacks.run('on_train_batch_start')
ni = i + nb * epoch # number integrated batches (since train start)
imgs = imgs.to(device, non_blocking=True).float() / 255 # uint8 to float32, 0-255 to 0.0-1.0
# Warmup
if ni <= nw:
xi = [0, nw] # x interp
# compute_loss.gr = np.interp(ni, xi, [0.0, 1.0]) # iou loss ratio (obj_loss = 1.0 or iou)
accumulate = max(1, np.interp(ni, xi, [1, nbs / batch_size]).round())
for j, x in enumerate(optimizer.param_groups):
# bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
x['lr'] = np.interp(ni, xi, [hyp['warmup_bias_lr'] if j == 0 else 0.0, x['initial_lr'] * lf(epoch)])
if 'momentum' in x:
x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])
# Multi-scale
if opt.multi_scale:
sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs # size
sf = sz / max(imgs.shape[2:]) # scale factor
if sf != 1:
ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]] # new shape (stretched to gs-multiple)
imgs = nn.functional.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)
# Forward
with torch.cuda.amp.autocast(amp):
pred = model(imgs) # forward
loss, loss_items = compute_loss(pred, targets.to(device), masks=masks.to(device).float())
if RANK != -1:
loss *= WORLD_SIZE # gradient averaged between devices in DDP mode
if opt.quad:
loss *= 4.
# Backward
scaler.scale(loss).backward()
# Optimize - https://pytorch.org/docs/master/notes/amp_examples.html
if ni - last_opt_step >= accumulate:
scaler.unscale_(optimizer) # unscale gradients
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=10.0) # clip gradients
scaler.step(optimizer) # optimizer.step
scaler.update()
optimizer.zero_grad()
if ema:
ema.update(model)
last_opt_step = ni
# Log
if RANK in {-1, 0}:
mloss = (mloss * i + loss_items) / (i + 1) # update mean losses
mem = f'{torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0:.3g}G' # (GB)
pbar.set_description(('%11s' * 2 + '%11.4g' * 6) %
(f'{epoch}/{epochs - 1}', mem, *mloss, targets.shape[0], imgs.shape[-1]))
# callbacks.run('on_train_batch_end', model, ni, imgs, targets, paths)
# if callbacks.stop_training:
# return
# Mosaic plots
if plots:
if ni < 3:
plot_images_and_masks(imgs, targets, masks, paths, save_dir / f"train_batch{ni}.jpg")
if ni == 10:
files = sorted(save_dir.glob('train*.jpg'))
logger.log_images(files, "Mosaics", epoch)
# end batch ------------------------------------------------------------------------------------------------
# Scheduler
lr = [x['lr'] for x in optimizer.param_groups] # for loggers
scheduler.step()
if RANK in {-1, 0}:
# mAP
# callbacks.run('on_train_epoch_end', epoch=epoch)
ema.update_attr(model, include=['yaml', 'nc', 'hyp', 'names', 'stride', 'class_weights'])
final_epoch = (epoch + 1 == epochs) or stopper.possible_stop
if not noval or final_epoch: # Calculate mAP
results, maps, _ = validate.run(data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
half=amp,
model=ema.ema,
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
plots=False,
callbacks=callbacks,
compute_loss=compute_loss,
mask_downsample_ratio=mask_ratio,
overlap=overlap)
# Update best mAP
fi = fitness(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
stop = stopper(epoch=epoch, fitness=fi) # early stop check
if fi > best_fitness:
best_fitness = fi
log_vals = list(mloss) + list(results) + lr
# callbacks.run('on_fit_epoch_end', log_vals, epoch, best_fitness, fi)
# Log val metrics and media
metrics_dict = dict(zip(KEYS, log_vals))
logger.log_metrics(metrics_dict, epoch)
# Save model
if (not nosave) or (final_epoch and not evolve): # if save
ckpt = {
'epoch': epoch,
'best_fitness': best_fitness,
'model': deepcopy(de_parallel(model)).half(),
'ema': deepcopy(ema.ema).half(),
'updates': ema.updates,
'optimizer': optimizer.state_dict(),
'opt': vars(opt),
'git': GIT_INFO, # {remote, branch, commit} if a git repo
'date': datetime.now().isoformat()}
# Save last, best and delete
torch.save(ckpt, last)
if best_fitness == fi:
torch.save(ckpt, best)
if opt.save_period > 0 and epoch % opt.save_period == 0:
torch.save(ckpt, w / f'epoch{epoch}.pt')
logger.log_model(w / f'epoch{epoch}.pt')
del ckpt
# callbacks.run('on_model_save', last, epoch, final_epoch, best_fitness, fi)
# EarlyStopping
if RANK != -1: # if DDP training
broadcast_list = [stop if RANK == 0 else None]
dist.broadcast_object_list(broadcast_list, 0) # broadcast 'stop' to all ranks
if RANK != 0:
stop = broadcast_list[0]
if stop:
break # must break all DDP ranks
# end epoch ----------------------------------------------------------------------------------------------------
# end training -----------------------------------------------------------------------------------------------------
if RANK in {-1, 0}:
LOGGER.info(f'\n{epoch - start_epoch + 1} epochs completed in {(time.time() - t0) / 3600:.3f} hours.')
for f in last, best:
if f.exists():
strip_optimizer(f) # strip optimizers
if f is best:
LOGGER.info(f'\nValidating {f}...')
results, _, _ = validate.run(
data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
model=attempt_load(f, device).half(),
iou_thres=0.65 if is_coco else 0.60, # best pycocotools at iou 0.65
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
save_json=is_coco,
verbose=True,
plots=plots,
callbacks=callbacks,
compute_loss=compute_loss,
mask_downsample_ratio=mask_ratio,
overlap=overlap) # val best model with plots
if is_coco:
# callbacks.run('on_fit_epoch_end', list(mloss) + list(results) + lr, epoch, best_fitness, fi)
metrics_dict = dict(zip(KEYS, list(mloss) + list(results) + lr))
logger.log_metrics(metrics_dict, epoch)
# callbacks.run('on_train_end', last, best, epoch, results)
# on train end callback using genericLogger
logger.log_metrics(dict(zip(KEYS[4:16], results)), epochs)
if not opt.evolve:
logger.log_model(best, epoch)
if plots:
plot_results_with_masks(file=save_dir / 'results.csv') # save results.png
files = ['results.png', 'confusion_matrix.png', *(f'{x}_curve.png' for x in ('F1', 'PR', 'P', 'R'))]
files = [(save_dir / f) for f in files if (save_dir / f).exists()] # filter
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}")
logger.log_images(files, "Results", epoch + 1)
logger.log_images(sorted(save_dir.glob('val*.jpg')), "Validation", epoch + 1)
torch.cuda.empty_cache()
return results
def parse_opt(known=False):
parser = argparse.ArgumentParser()
parser.add_argument('--weights', type=str, default=ROOT / 'yolo-seg.pt', help='initial weights path')
parser.add_argument('--cfg', type=str, default='', help='model.yaml path')
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128-seg.yaml', help='dataset.yaml path')
parser.add_argument('--hyp', type=str, default=ROOT / 'data/hyps/hyp.scratch-low.yaml', help='hyperparameters path')
parser.add_argument('--epochs', type=int, default=100, help='total training epochs')
parser.add_argument('--batch-size', type=int, default=16, help='total batch size for all GPUs, -1 for autobatch')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='train, val image size (pixels)')
parser.add_argument('--rect', action='store_true', help='rectangular training')
parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
parser.add_argument('--noval', action='store_true', help='only validate final epoch')
parser.add_argument('--noautoanchor', action='store_true', help='disable AutoAnchor')
parser.add_argument('--noplots', action='store_true', help='save no plot files')
parser.add_argument('--evolve', type=int, nargs='?', const=300, help='evolve hyperparameters for x generations')
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
parser.add_argument('--cache', type=str, nargs='?', const='ram', help='image --cache ram/disk')
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
parser.add_argument('--single-cls', action='store_true', help='train multi-class data as single-class')
parser.add_argument('--optimizer', type=str, choices=['SGD', 'Adam', 'AdamW', 'LION'], default='SGD', help='optimizer')
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
parser.add_argument('--project', default=ROOT / 'runs/train-seg', help='save to project/name')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--quad', action='store_true', help='quad dataloader')
parser.add_argument('--cos-lr', action='store_true', help='cosine LR scheduler')
parser.add_argument('--label-smoothing', type=float, default=0.0, help='Label smoothing epsilon')
parser.add_argument('--patience', type=int, default=100, help='EarlyStopping patience (epochs without improvement)')
parser.add_argument('--freeze', nargs='+', type=int, default=[0], help='Freeze layers: backbone=10, first3=0 1 2')
parser.add_argument('--save-period', type=int, default=-1, help='Save checkpoint every x epochs (disabled if < 1)')
parser.add_argument('--seed', type=int, default=0, help='Global training seed')
parser.add_argument('--local_rank', type=int, default=-1, help='Automatic DDP Multi-GPU argument, do not modify')
parser.add_argument('--close-mosaic', type=int, default=0, help='Experimental')
# Instance Segmentation Args
parser.add_argument('--mask-ratio', type=int, default=4, help='Downsample the truth masks to saving memory')
parser.add_argument('--no-overlap', action='store_true', help='Overlap masks train faster at slightly less mAP')
return parser.parse_known_args()[0] if known else parser.parse_args()
def main(opt, callbacks=Callbacks()):
# Checks
if RANK in {-1, 0}:
print_args(vars(opt))
#check_git_status()
#check_requirements()
# Resume
if opt.resume and not opt.evolve: # resume from specified or most recent last.pt
last = Path(check_file(opt.resume) if isinstance(opt.resume, str) else get_latest_run())
opt_yaml = last.parent.parent / 'opt.yaml' # train options yaml
opt_data = opt.data # original dataset
if opt_yaml.is_file():
with open(opt_yaml, errors='ignore') as f:
d = yaml.safe_load(f)
else:
d = torch.load(last, map_location='cpu')['opt']
opt = argparse.Namespace(**d) # replace
opt.cfg, opt.weights, opt.resume = '', str(last), True # reinstate
if is_url(opt_data):
opt.data = check_file(opt_data) # avoid HUB resume auth timeout
else:
opt.data, opt.cfg, opt.hyp, opt.weights, opt.project = \
check_file(opt.data), check_yaml(opt.cfg), check_yaml(opt.hyp), str(opt.weights), str(opt.project) # checks
assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
if opt.evolve:
if opt.project == str(ROOT / 'runs/train'): # if default project name, rename to runs/evolve
opt.project = str(ROOT / 'runs/evolve')
opt.exist_ok, opt.resume = opt.resume, False # pass resume to exist_ok and disable resume
if opt.name == 'cfg':
opt.name = Path(opt.cfg).stem # use model.yaml as name
opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok))
# DDP mode
device = select_device(opt.device, batch_size=opt.batch_size)
if LOCAL_RANK != -1:
msg = 'is not compatible with YOLO Multi-GPU DDP training'
assert not opt.image_weights, f'--image-weights {msg}'
assert not opt.evolve, f'--evolve {msg}'
assert opt.batch_size != -1, f'AutoBatch with --batch-size -1 {msg}, please pass a valid --batch-size'
assert opt.batch_size % WORLD_SIZE == 0, f'--batch-size {opt.batch_size} must be multiple of WORLD_SIZE'
assert torch.cuda.device_count() > LOCAL_RANK, 'insufficient CUDA devices for DDP command'
torch.cuda.set_device(LOCAL_RANK)
device = torch.device('cuda', LOCAL_RANK)
dist.init_process_group(backend="nccl" if dist.is_nccl_available() else "gloo")
# Train
if not opt.evolve:
train(opt.hyp, opt, device, callbacks)
# Evolve hyperparameters (optional)
else:
# Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
meta = {
'lr0': (1, 1e-5, 1e-1), # initial learning rate (SGD=1E-2, Adam=1E-3)
'lrf': (1, 0.01, 1.0), # final OneCycleLR learning rate (lr0 * lrf)
'momentum': (0.3, 0.6, 0.98), # SGD momentum/Adam beta1
'weight_decay': (1, 0.0, 0.001), # optimizer weight decay
'warmup_epochs': (1, 0.0, 5.0), # warmup epochs (fractions ok)
'warmup_momentum': (1, 0.0, 0.95), # warmup initial momentum
'warmup_bias_lr': (1, 0.0, 0.2), # warmup initial bias lr
'box': (1, 0.02, 0.2), # box loss gain
'cls': (1, 0.2, 4.0), # cls loss gain
'cls_pw': (1, 0.5, 2.0), # cls BCELoss positive_weight
'obj': (1, 0.2, 4.0), # obj loss gain (scale with pixels)
'obj_pw': (1, 0.5, 2.0), # obj BCELoss positive_weight
'iou_t': (0, 0.1, 0.7), # IoU training threshold
'anchor_t': (1, 2.0, 8.0), # anchor-multiple threshold
'anchors': (2, 2.0, 10.0), # anchors per output grid (0 to ignore)
'fl_gamma': (0, 0.0, 2.0), # focal loss gamma (efficientDet default gamma=1.5)
'hsv_h': (1, 0.0, 0.1), # image HSV-Hue augmentation (fraction)
'hsv_s': (1, 0.0, 0.9), # image HSV-Saturation augmentation (fraction)
'hsv_v': (1, 0.0, 0.9), # image HSV-Value augmentation (fraction)
'degrees': (1, 0.0, 45.0), # image rotation (+/- deg)
'translate': (1, 0.0, 0.9), # image translation (+/- fraction)
'scale': (1, 0.0, 0.9), # image scale (+/- gain)
'shear': (1, 0.0, 10.0), # image shear (+/- deg)
'perspective': (0, 0.0, 0.001), # image perspective (+/- fraction), range 0-0.001
'flipud': (1, 0.0, 1.0), # image flip up-down (probability)
'fliplr': (0, 0.0, 1.0), # image flip left-right (probability)
'mosaic': (1, 0.0, 1.0), # image mixup (probability)
'mixup': (1, 0.0, 1.0), # image mixup (probability)
'copy_paste': (1, 0.0, 1.0)} # segment copy-paste (probability)
with open(opt.hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
if 'anchors' not in hyp: # anchors commented in hyp.yaml
hyp['anchors'] = 3
if opt.noautoanchor:
del hyp['anchors'], meta['anchors']
opt.noval, opt.nosave, save_dir = True, True, Path(opt.save_dir) # only val/save final epoch
# ei = [isinstance(x, (int, float)) for x in hyp.values()] # evolvable indices
evolve_yaml, evolve_csv = save_dir / 'hyp_evolve.yaml', save_dir / 'evolve.csv'
if opt.bucket:
os.system(f'gsutil cp gs://{opt.bucket}/evolve.csv {evolve_csv}') # download evolve.csv if exists
for _ in range(opt.evolve): # generations to evolve
if evolve_csv.exists(): # if evolve.csv exists: select best hyps and mutate
# Select parent(s)
parent = 'single' # parent selection method: 'single' or 'weighted'
x = np.loadtxt(evolve_csv, ndmin=2, delimiter=',', skiprows=1)
n = min(5, len(x)) # number of previous results to consider
x = x[np.argsort(-fitness(x))][:n] # top n mutations
w = fitness(x) - fitness(x).min() + 1E-6 # weights (sum > 0)
if parent == 'single' or len(x) == 1:
# x = x[random.randint(0, n - 1)] # random selection
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
elif parent == 'weighted':
x = (x * w.reshape(n, 1)).sum(0) / w.sum() # weighted combination
# Mutate
mp, s = 0.8, 0.2 # mutation probability, sigma
npr = np.random
npr.seed(int(time.time()))
g = np.array([meta[k][0] for k in hyp.keys()]) # gains 0-1
ng = len(meta)
v = np.ones(ng)
while all(v == 1): # mutate until a change occurs (prevent duplicates)
v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
for i, k in enumerate(hyp.keys()): # plt.hist(v.ravel(), 300)
hyp[k] = float(x[i + 7] * v[i]) # mutate
# Constrain to limits
for k, v in meta.items():
hyp[k] = max(hyp[k], v[1]) # lower limit
hyp[k] = min(hyp[k], v[2]) # upper limit
hyp[k] = round(hyp[k], 5) # significant digits
# Train mutation
results = train(hyp.copy(), opt, device, callbacks)
callbacks = Callbacks()
# Write mutation results
print_mutation(KEYS, results, hyp.copy(), save_dir, opt.bucket)
# Plot results
plot_evolve(evolve_csv)
LOGGER.info(f'Hyperparameter evolution finished {opt.evolve} generations\n'
f"Results saved to {colorstr('bold', save_dir)}\n"
f'Usage example: $ python train.py --hyp {evolve_yaml}')
def run(**kwargs):
# Usage: import train; train.run(data='coco128.yaml', imgsz=320, weights='yolo.pt')
opt = parse_opt(True)
for k, v in kwargs.items():
setattr(opt, k, v)
main(opt)
return opt
if __name__ == "__main__":
opt = parse_opt()
main(opt)
segment/val.py 0 → 100644
View file @ f969ca34
import argparse
import json
import os
import sys
from multiprocessing.pool import ThreadPool
from pathlib import Path
import numpy as np
import torch
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
import torch.nn.functional as F
from models.common import DetectMultiBackend
from models.yolo import SegmentationModel
from utils.callbacks import Callbacks
from utils.general import (LOGGER, NUM_THREADS, TQDM_BAR_FORMAT, Profile, check_dataset, check_img_size,
check_requirements, check_yaml, coco80_to_coco91_class, colorstr, increment_path,
non_max_suppression, print_args, scale_boxes, xywh2xyxy, xyxy2xywh)
from utils.metrics import ConfusionMatrix, box_iou
from utils.plots import output_to_target, plot_val_study
from utils.segment.dataloaders import create_dataloader
from utils.segment.general import mask_iou, process_mask, process_mask_upsample, scale_image
from utils.segment.metrics import Metrics, ap_per_class_box_and_mask
from utils.segment.plots import plot_images_and_masks
from utils.torch_utils import de_parallel, select_device, smart_inference_mode
def save_one_txt(predn, save_conf, shape, file):
# Save one txt result
gn = torch.tensor(shape)[[1, 0, 1, 0]] # normalization gain whwh
for *xyxy, conf, cls in predn.tolist():
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format
with open(file, 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
def save_one_json(predn, jdict, path, class_map, pred_masks):
# Save one JSON result {"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236}
from pycocotools.mask import encode
def single_encode(x):
rle = encode(np.asarray(x[:, :, None], order="F", dtype="uint8"))[0]
rle["counts"] = rle["counts"].decode("utf-8")
return rle
image_id = int(path.stem) if path.stem.isnumeric() else path.stem
box = xyxy2xywh(predn[:, :4]) # xywh
box[:, :2] -= box[:, 2:] / 2 # xy center to top-left corner
pred_masks = np.transpose(pred_masks, (2, 0, 1))
with ThreadPool(NUM_THREADS) as pool:
rles = pool.map(single_encode, pred_masks)
for i, (p, b) in enumerate(zip(predn.tolist(), box.tolist())):
jdict.append({
'image_id': image_id,
'category_id': class_map[int(p[5])],
'bbox': [round(x, 3) for x in b],
'score': round(p[4], 5),
'segmentation': rles[i]})
def process_batch(detections, labels, iouv, pred_masks=None, gt_masks=None, overlap=False, masks=False):
"""
Return correct prediction matrix
Arguments:
detections (array[N, 6]), x1, y1, x2, y2, conf, class
labels (array[M, 5]), class, x1, y1, x2, y2
Returns:
correct (array[N, 10]), for 10 IoU levels
"""
if masks:
if overlap:
nl = len(labels)
index = torch.arange(nl, device=gt_masks.device).view(nl, 1, 1) + 1
gt_masks = gt_masks.repeat(nl, 1, 1) # shape(1,640,640) -> (n,640,640)
gt_masks = torch.where(gt_masks == index, 1.0, 0.0)
if gt_masks.shape[1:] != pred_masks.shape[1:]:
gt_masks = F.interpolate(gt_masks[None], pred_masks.shape[1:], mode="bilinear", align_corners=False)[0]
gt_masks = gt_masks.gt_(0.5)
iou = mask_iou(gt_masks.view(gt_masks.shape[0], -1), pred_masks.view(pred_masks.shape[0], -1))
else: # boxes
iou = box_iou(labels[:, 1:], detections[:, :4])
correct = np.zeros((detections.shape[0], iouv.shape[0])).astype(bool)
correct_class = labels[:, 0:1] == detections[:, 5]
for i in range(len(iouv)):
x = torch.where((iou >= iouv[i]) & correct_class) # IoU > threshold and classes match
if x[0].shape[0]:
matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy() # [label, detect, iou]
if x[0].shape[0] > 1:
matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
# matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
correct[matches[:, 1].astype(int), i] = True
return torch.tensor(correct, dtype=torch.bool, device=iouv.device)
@smart_inference_mode()
def run(
data,
weights=None, # model.pt path(s)
batch_size=32, # batch size
imgsz=640, # inference size (pixels)
conf_thres=0.001, # confidence threshold
iou_thres=0.6, # NMS IoU threshold
max_det=300, # maximum detections per image
task='val', # train, val, test, speed or study
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
workers=8, # max dataloader workers (per RANK in DDP mode)
single_cls=False, # treat as single-class dataset
augment=False, # augmented inference
verbose=False, # verbose output
save_txt=False, # save results to *.txt
save_hybrid=False, # save label+prediction hybrid results to *.txt
save_conf=False, # save confidences in --save-txt labels
save_json=False, # save a COCO-JSON results file
project=ROOT / 'runs/val-seg', # save to project/name
name='exp', # save to project/name
exist_ok=False, # existing project/name ok, do not increment
half=True, # use FP16 half-precision inference
dnn=False, # use OpenCV DNN for ONNX inference
model=None,
dataloader=None,
save_dir=Path(''),
plots=True,
overlap=False,
mask_downsample_ratio=1,
compute_loss=None,
callbacks=Callbacks(),
):
if save_json:
check_requirements(['pycocotools'])
process = process_mask_upsample # more accurate
else:
process = process_mask # faster
# Initialize/load model and set device
training = model is not None
if training: # called by train.py
device, pt, jit, engine = next(model.parameters()).device, True, False, False # get model device, PyTorch model
half &= device.type != 'cpu' # half precision only supported on CUDA
model.half() if half else model.float()
nm = de_parallel(model).model[-1].nm # number of masks
else: # called directly
device = select_device(device, batch_size=batch_size)
# Directories
save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
# Load model
model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
stride, pt, jit, engine = model.stride, model.pt, model.jit, model.engine
imgsz = check_img_size(imgsz, s=stride) # check image size
half = model.fp16 # FP16 supported on limited backends with CUDA
nm = de_parallel(model).model.model[-1].nm if isinstance(model, SegmentationModel) else 32 # number of masks
if engine:
batch_size = model.batch_size
else:
device = model.device
if not (pt or jit):
batch_size = 1 # export.py models default to batch-size 1
LOGGER.info(f'Forcing --batch-size 1 square inference (1,3,{imgsz},{imgsz}) for non-PyTorch models')
# Data
data = check_dataset(data) # check
# Configure
model.eval()
cuda = device.type != 'cpu'
#is_coco = isinstance(data.get('val'), str) and data['val'].endswith(f'coco{os.sep}val2017.txt') # COCO dataset
is_coco = isinstance(data.get('val'), str) and data['val'].endswith(f'val2017.txt') # COCO dataset
nc = 1 if single_cls else int(data['nc']) # number of classes
iouv = torch.linspace(0.5, 0.95, 10, device=device) # iou vector for mAP@0.5:0.95
niou = iouv.numel()
# Dataloader
if not training:
if pt and not single_cls: # check --weights are trained on --data
ncm = model.model.nc
assert ncm == nc, f'{weights} ({ncm} classes) trained on different --data than what you passed ({nc} ' \
f'classes). Pass correct combination of --weights and --data that are trained together.'
model.warmup(imgsz=(1 if pt else batch_size, 3, imgsz, imgsz)) # warmup
pad, rect = (0.0, False) if task == 'speed' else (0.5, pt) # square inference for benchmarks
task = task if task in ('train', 'val', 'test') else 'val' # path to train/val/test images
dataloader = create_dataloader(data[task],
imgsz,
batch_size,
stride,
single_cls,
pad=pad,
rect=rect,
workers=workers,
prefix=colorstr(f'{task}: '),
overlap_mask=overlap,
mask_downsample_ratio=mask_downsample_ratio)[0]
seen = 0
confusion_matrix = ConfusionMatrix(nc=nc)
names = model.names if hasattr(model, 'names') else model.module.names # get class names
if isinstance(names, (list, tuple)): # old format
names = dict(enumerate(names))
class_map = coco80_to_coco91_class() if is_coco else list(range(1000))
s = ('%22s' + '%11s' * 10) % ('Class', 'Images', 'Instances', 'Box(P', "R", "mAP50", "mAP50-95)", "Mask(P", "R",
"mAP50", "mAP50-95)")
dt = Profile(), Profile(), Profile()
metrics = Metrics()
loss = torch.zeros(4, device=device)
jdict, stats = [], []
# callbacks.run('on_val_start')
pbar = tqdm(dataloader, desc=s, bar_format=TQDM_BAR_FORMAT) # progress bar
for batch_i, (im, targets, paths, shapes, masks) in enumerate(pbar):
# callbacks.run('on_val_batch_start')
with dt[0]:
if cuda:
im = im.to(device, non_blocking=True)
targets = targets.to(device)
masks = masks.to(device)
masks = masks.float()
im = im.half() if half else im.float() # uint8 to fp16/32
im /= 255 # 0 - 255 to 0.0 - 1.0
nb, _, height, width = im.shape # batch size, channels, height, width
# Inference
with dt[1]:
preds, train_out = model(im)# if compute_loss else (*model(im, augment=augment)[:2], None)
#train_out, preds, protos = p if len(p) == 3 else p[1]
#preds = p
#train_out = p[1][0] if len(p[1]) == 3 else p[0]
protos = train_out[-1]
#print(preds.shape)
#print(train_out[0].shape)
#print(train_out[1].shape)
#print(train_out[2].shape)
# Loss
if compute_loss:
loss += compute_loss(train_out, targets, masks)[1] # box, obj, cls
# NMS
targets[:, 2:] *= torch.tensor((width, height, width, height), device=device) # to pixels
lb = [targets[targets[:, 0] == i, 1:] for i in range(nb)] if save_hybrid else [] # for autolabelling
with dt[2]:
preds = non_max_suppression(preds,
conf_thres,
iou_thres,
labels=lb,
multi_label=True,
agnostic=single_cls,
max_det=max_det,
nm=nm)
# Metrics
plot_masks = [] # masks for plotting
for si, (pred, proto) in enumerate(zip(preds, protos)):
labels = targets[targets[:, 0] == si, 1:]
nl, npr = labels.shape[0], pred.shape[0] # number of labels, predictions
path, shape = Path(paths[si]), shapes[si][0]
correct_masks = torch.zeros(npr, niou, dtype=torch.bool, device=device) # init
correct_bboxes = torch.zeros(npr, niou, dtype=torch.bool, device=device) # init
seen += 1
if npr == 0:
if nl:
stats.append((correct_masks, correct_bboxes, *torch.zeros((2, 0), device=device), labels[:, 0]))
if plots:
confusion_matrix.process_batch(detections=None, labels=labels[:, 0])
continue
# Masks
midx = [si] if overlap else targets[:, 0] == si
gt_masks = masks[midx]
pred_masks = process(proto, pred[:, 6:], pred[:, :4], shape=im[si].shape[1:])
# Predictions
if single_cls:
pred[:, 5] = 0
predn = pred.clone()
scale_boxes(im[si].shape[1:], predn[:, :4], shape, shapes[si][1]) # native-space pred
# Evaluate
if nl:
tbox = xywh2xyxy(labels[:, 1:5]) # target boxes
scale_boxes(im[si].shape[1:], tbox, shape, shapes[si][1]) # native-space labels
labelsn = torch.cat((labels[:, 0:1], tbox), 1) # native-space labels
correct_bboxes = process_batch(predn, labelsn, iouv)
correct_masks = process_batch(predn, labelsn, iouv, pred_masks, gt_masks, overlap=overlap, masks=True)
if plots:
confusion_matrix.process_batch(predn, labelsn)
stats.append((correct_masks, correct_bboxes, pred[:, 4], pred[:, 5], labels[:, 0])) # (conf, pcls, tcls)
pred_masks = torch.as_tensor(pred_masks, dtype=torch.uint8)
if plots and batch_i < 3:
plot_masks.append(pred_masks[:15].cpu()) # filter top 15 to plot
# Save/log
if save_txt:
save_one_txt(predn, save_conf, shape, file=save_dir / 'labels' / f'{path.stem}.txt')
if save_json:
pred_masks = scale_image(im[si].shape[1:],
pred_masks.permute(1, 2, 0).contiguous().cpu().numpy(), shape, shapes[si][1])
save_one_json(predn, jdict, path, class_map, pred_masks) # append to COCO-JSON dictionary
# callbacks.run('on_val_image_end', pred, predn, path, names, im[si])
# Plot images
if plots and batch_i < 3:
if len(plot_masks):
plot_masks = torch.cat(plot_masks, dim=0)
plot_images_and_masks(im, targets, masks, paths, save_dir / f'val_batch{batch_i}_labels.jpg', names)
plot_images_and_masks(im, output_to_target(preds, max_det=15), plot_masks, paths,
save_dir / f'val_batch{batch_i}_pred.jpg', names) # pred
# callbacks.run('on_val_batch_end')
# Compute metrics
stats = [torch.cat(x, 0).cpu().numpy() for x in zip(*stats)] # to numpy
if len(stats) and stats[0].any():
results = ap_per_class_box_and_mask(*stats, plot=plots, save_dir=save_dir, names=names)
metrics.update(results)
nt = np.bincount(stats[4].astype(int), minlength=nc) # number of targets per class
# Print results
pf = '%22s' + '%11i' * 2 + '%11.3g' * 8 # print format
LOGGER.info(pf % ("all", seen, nt.sum(), *metrics.mean_results()))
if nt.sum() == 0:
LOGGER.warning(f'WARNING ⚠️ no labels found in {task} set, can not compute metrics without labels')
# Print results per class
if (verbose or (nc < 50 and not training)) and nc > 1 and len(stats):
for i, c in enumerate(metrics.ap_class_index):
LOGGER.info(pf % (names[c], seen, nt[c], *metrics.class_result(i)))
# Print speeds
t = tuple(x.t / seen * 1E3 for x in dt) # speeds per image
if not training:
shape = (batch_size, 3, imgsz, imgsz)
LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {shape}' % t)
# Plots
if plots:
confusion_matrix.plot(save_dir=save_dir, names=list(names.values()))
# callbacks.run('on_val_end')
mp_bbox, mr_bbox, map50_bbox, map_bbox, mp_mask, mr_mask, map50_mask, map_mask = metrics.mean_results()
# Save JSON
if save_json and len(jdict):
w = Path(weights[0] if isinstance(weights, list) else weights).stem if weights is not None else '' # weights
anno_json = str(Path(data.get('path', '../coco')) / 'annotations/instances_val2017.json') # annotations json
pred_json = str(save_dir / f"{w}_predictions.json") # predictions json
LOGGER.info(f'\nEvaluating pycocotools mAP... saving {pred_json}...')
with open(pred_json, 'w') as f:
json.dump(jdict, f)
try: # https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval
anno = COCO(anno_json) # init annotations api
pred = anno.loadRes(pred_json) # init predictions api
results = []
for eval in COCOeval(anno, pred, 'bbox'), COCOeval(anno, pred, 'segm'):
if is_coco:
eval.params.imgIds = [int(Path(x).stem) for x in dataloader.dataset.im_files] # img ID to evaluate
eval.evaluate()
eval.accumulate()
eval.summarize()
results.extend(eval.stats[:2]) # update results (mAP@0.5:0.95, mAP@0.5)
map_bbox, map50_bbox, map_mask, map50_mask = results
except Exception as e:
LOGGER.info(f'pycocotools unable to run: {e}')
# Return results
model.float() # for training
if not training:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
final_metric = mp_bbox, mr_bbox, map50_bbox, map_bbox, mp_mask, mr_mask, map50_mask, map_mask
return (*final_metric, *(loss.cpu() / len(dataloader)).tolist()), metrics.get_maps(nc), t
def parse_opt():
parser = argparse.ArgumentParser()
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128-seg.yaml', help='dataset.yaml path')
parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolo-seg.pt', help='model path(s)')
parser.add_argument('--batch-size', type=int, default=32, help='batch size')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='inference size (pixels)')
parser.add_argument('--conf-thres', type=float, default=0.001, help='confidence threshold')
parser.add_argument('--iou-thres', type=float, default=0.6, help='NMS IoU threshold')
parser.add_argument('--max-det', type=int, default=300, help='maximum detections per image')
parser.add_argument('--task', default='val', help='train, val, test, speed or study')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
parser.add_argument('--single-cls', action='store_true', help='treat as single-class dataset')
parser.add_argument('--augment', action='store_true', help='augmented inference')
parser.add_argument('--verbose', action='store_true', help='report mAP by class')
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
parser.add_argument('--save-hybrid', action='store_true', help='save label+prediction hybrid results to *.txt')
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
parser.add_argument('--save-json', action='store_true', help='save a COCO-JSON results file')
parser.add_argument('--project', default=ROOT / 'runs/val-seg', help='save results to project/name')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')
opt = parser.parse_args()
opt.data = check_yaml(opt.data) # check YAML
# opt.save_json |= opt.data.endswith('coco.yaml')
opt.save_txt |= opt.save_hybrid
print_args(vars(opt))
return opt
def main(opt):
#check_requirements(requirements=ROOT / 'requirements.txt', exclude=('tensorboard', 'thop'))
if opt.task in ('train', 'val', 'test'): # run normally
if opt.conf_thres > 0.001: # https://github.com/ultralytics/yolov5/issues/1466
LOGGER.warning(f'WARNING ⚠️ confidence threshold {opt.conf_thres} > 0.001 produces invalid results')
if opt.save_hybrid:
LOGGER.warning('WARNING ⚠️ --save-hybrid returns high mAP from hybrid labels, not from predictions alone')
run(**vars(opt))
else:
weights = opt.weights if isinstance(opt.weights, list) else [opt.weights]
opt.half = torch.cuda.is_available() and opt.device != 'cpu' # FP16 for fastest results
if opt.task == 'speed': # speed benchmarks
# python val.py --task speed --data coco.yaml --batch 1 --weights yolo.pt...
opt.conf_thres, opt.iou_thres, opt.save_json = 0.25, 0.45, False
for opt.weights in weights:
run(**vars(opt), plots=False)
elif opt.task == 'study': # speed vs mAP benchmarks
# python val.py --task study --data coco.yaml --iou 0.7 --weights yolo.pt...
for opt.weights in weights:
f = f'study_{Path(opt.data).stem}_{Path(opt.weights).stem}.txt' # filename to save to
x, y = list(range(256, 1536 + 128, 128)), [] # x axis (image sizes), y axis
for opt.imgsz in x: # img-size
LOGGER.info(f'\nRunning {f} --imgsz {opt.imgsz}...')
r, _, t = run(**vars(opt), plots=False)
y.append(r + t) # results and times
np.savetxt(f, y, fmt='%10.4g') # save
os.system('zip -r study.zip study_*.txt')
plot_val_study(x=x) # plot
if __name__ == "__main__":
opt = parse_opt()
main(opt)
segment/val_dual.py 0 → 100644
View file @ f969ca34
import argparse
import json
import os
import sys
from multiprocessing.pool import ThreadPool
from pathlib import Path
import numpy as np
import torch
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
import torch.nn.functional as F
from models.common import DetectMultiBackend
from models.yolo import SegmentationModel
from utils.callbacks import Callbacks
from utils.general import (LOGGER, NUM_THREADS, TQDM_BAR_FORMAT, Profile, check_dataset, check_img_size,
check_requirements, check_yaml, coco80_to_coco91_class, colorstr, increment_path,
non_max_suppression, print_args, scale_boxes, xywh2xyxy, xyxy2xywh)
from utils.metrics import ConfusionMatrix, box_iou
from utils.plots import output_to_target, plot_val_study
from utils.segment.dataloaders import create_dataloader
from utils.segment.general import mask_iou, process_mask, process_mask_upsample, scale_image
from utils.segment.metrics import Metrics, ap_per_class_box_and_mask
from utils.segment.plots import plot_images_and_masks
from utils.torch_utils import de_parallel, select_device, smart_inference_mode
def save_one_txt(predn, save_conf, shape, file):
# Save one txt result
gn = torch.tensor(shape)[[1, 0, 1, 0]] # normalization gain whwh
for *xyxy, conf, cls in predn.tolist():
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format
with open(file, 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
def save_one_json(predn, jdict, path, class_map, pred_masks):
# Save one JSON result {"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236}
from pycocotools.mask import encode
def single_encode(x):
rle = encode(np.asarray(x[:, :, None], order="F", dtype="uint8"))[0]
rle["counts"] = rle["counts"].decode("utf-8")
return rle
image_id = int(path.stem) if path.stem.isnumeric() else path.stem
box = xyxy2xywh(predn[:, :4]) # xywh
box[:, :2] -= box[:, 2:] / 2 # xy center to top-left corner
pred_masks = np.transpose(pred_masks, (2, 0, 1))
with ThreadPool(NUM_THREADS) as pool:
rles = pool.map(single_encode, pred_masks)
for i, (p, b) in enumerate(zip(predn.tolist(), box.tolist())):
jdict.append({
'image_id': image_id,
'category_id': class_map[int(p[5])],
'bbox': [round(x, 3) for x in b],
'score': round(p[4], 5),
'segmentation': rles[i]})
def process_batch(detections, labels, iouv, pred_masks=None, gt_masks=None, overlap=False, masks=False):
"""
Return correct prediction matrix
Arguments:
detections (array[N, 6]), x1, y1, x2, y2, conf, class
labels (array[M, 5]), class, x1, y1, x2, y2
Returns:
correct (array[N, 10]), for 10 IoU levels
"""
if masks:
if overlap:
nl = len(labels)
index = torch.arange(nl, device=gt_masks.device).view(nl, 1, 1) + 1
gt_masks = gt_masks.repeat(nl, 1, 1) # shape(1,640,640) -> (n,640,640)
gt_masks = torch.where(gt_masks == index, 1.0, 0.0)
if gt_masks.shape[1:] != pred_masks.shape[1:]:
gt_masks = F.interpolate(gt_masks[None], pred_masks.shape[1:], mode="bilinear", align_corners=False)[0]
gt_masks = gt_masks.gt_(0.5)
iou = mask_iou(gt_masks.view(gt_masks.shape[0], -1), pred_masks.view(pred_masks.shape[0], -1))
else: # boxes
iou = box_iou(labels[:, 1:], detections[:, :4])
correct = np.zeros((detections.shape[0], iouv.shape[0])).astype(bool)
correct_class = labels[:, 0:1] == detections[:, 5]
for i in range(len(iouv)):
x = torch.where((iou >= iouv[i]) & correct_class) # IoU > threshold and classes match
if x[0].shape[0]:
matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy() # [label, detect, iou]
if x[0].shape[0] > 1:
matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
# matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
correct[matches[:, 1].astype(int), i] = True
return torch.tensor(correct, dtype=torch.bool, device=iouv.device)
@smart_inference_mode()
def run(
data,
weights=None, # model.pt path(s)
batch_size=32, # batch size
imgsz=640, # inference size (pixels)
conf_thres=0.001, # confidence threshold
iou_thres=0.6, # NMS IoU threshold
max_det=300, # maximum detections per image
task='val', # train, val, test, speed or study
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
workers=8, # max dataloader workers (per RANK in DDP mode)
single_cls=False, # treat as single-class dataset
augment=False, # augmented inference
verbose=False, # verbose output
save_txt=False, # save results to *.txt
save_hybrid=False, # save label+prediction hybrid results to *.txt
save_conf=False, # save confidences in --save-txt labels
save_json=False, # save a COCO-JSON results file
project=ROOT / 'runs/val-seg', # save to project/name
name='exp', # save to project/name
exist_ok=False, # existing project/name ok, do not increment
half=True, # use FP16 half-precision inference
dnn=False, # use OpenCV DNN for ONNX inference
model=None,
dataloader=None,
save_dir=Path(''),
plots=True,
overlap=False,
mask_downsample_ratio=1,
compute_loss=None,
callbacks=Callbacks(),
):
if save_json:
check_requirements(['pycocotools'])
process = process_mask_upsample # more accurate
else:
process = process_mask # faster
# Initialize/load model and set device
training = model is not None
if training: # called by train.py
device, pt, jit, engine = next(model.parameters()).device, True, False, False # get model device, PyTorch model
half &= device.type != 'cpu' # half precision only supported on CUDA
model.half() if half else model.float()
nm = de_parallel(model).model[-1].nm # number of masks
else: # called directly
device = select_device(device, batch_size=batch_size)
# Directories
save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
# Load model
model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
stride, pt, jit, engine = model.stride, model.pt, model.jit, model.engine
imgsz = check_img_size(imgsz, s=stride) # check image size
half = model.fp16 # FP16 supported on limited backends with CUDA
nm = de_parallel(model).model.model[-1].nm if isinstance(model, SegmentationModel) else 32 # number of masks
if engine:
batch_size = model.batch_size
else:
device = model.device
if not (pt or jit):
batch_size = 1 # export.py models default to batch-size 1
LOGGER.info(f'Forcing --batch-size 1 square inference (1,3,{imgsz},{imgsz}) for non-PyTorch models')
# Data
data = check_dataset(data) # check
# Configure
model.eval()
cuda = device.type != 'cpu'
#is_coco = isinstance(data.get('val'), str) and data['val'].endswith(f'coco{os.sep}val2017.txt') # COCO dataset
is_coco = isinstance(data.get('val'), str) and data['val'].endswith(f'val2017.txt') # COCO dataset
nc = 1 if single_cls else int(data['nc']) # number of classes
iouv = torch.linspace(0.5, 0.95, 10, device=device) # iou vector for mAP@0.5:0.95
niou = iouv.numel()
# Dataloader
if not training:
if pt and not single_cls: # check --weights are trained on --data
ncm = model.model.nc
assert ncm == nc, f'{weights} ({ncm} classes) trained on different --data than what you passed ({nc} ' \
f'classes). Pass correct combination of --weights and --data that are trained together.'
model.warmup(imgsz=(1 if pt else batch_size, 3, imgsz, imgsz)) # warmup
pad, rect = (0.0, False) if task == 'speed' else (0.5, pt) # square inference for benchmarks
task = task if task in ('train', 'val', 'test') else 'val' # path to train/val/test images
dataloader = create_dataloader(data[task],
imgsz,
batch_size,
stride,
single_cls,
pad=pad,
rect=rect,
workers=workers,
prefix=colorstr(f'{task}: '),
overlap_mask=overlap,
mask_downsample_ratio=mask_downsample_ratio)[0]
seen = 0
confusion_matrix = ConfusionMatrix(nc=nc)
names = model.names if hasattr(model, 'names') else model.module.names # get class names
if isinstance(names, (list, tuple)): # old format
names = dict(enumerate(names))
class_map = coco80_to_coco91_class() if is_coco else list(range(1000))
s = ('%22s' + '%11s' * 10) % ('Class', 'Images', 'Instances', 'Box(P', "R", "mAP50", "mAP50-95)", "Mask(P", "R",
"mAP50", "mAP50-95)")
dt = Profile(), Profile(), Profile()
metrics = Metrics()
loss = torch.zeros(4, device=device)
jdict, stats = [], []
# callbacks.run('on_val_start')
pbar = tqdm(dataloader, desc=s, bar_format=TQDM_BAR_FORMAT) # progress bar
for batch_i, (im, targets, paths, shapes, masks) in enumerate(pbar):
# callbacks.run('on_val_batch_start')
with dt[0]:
if cuda:
im = im.to(device, non_blocking=True)
targets = targets.to(device)
masks = masks.to(device)
masks = masks.float()
im = im.half() if half else im.float() # uint8 to fp16/32
im /= 255 # 0 - 255 to 0.0 - 1.0
nb, _, height, width = im.shape # batch size, channels, height, width
# Inference
with dt[1]:
preds, train_out = model(im)# if compute_loss else (*model(im, augment=augment)[:2], None)
#preds = preds[1]
#train_out, preds, protos = p if len(p) == 3 else p[1]
#preds = p
#train_out = p[1][0] if len(p[1]) == 3 else p[0]
protos = train_out[-1]
#print(preds.shape)
#print(train_out[0].shape)
#print(train_out[1].shape)
#print(train_out[2].shape)
# Loss
#if compute_loss:
# loss += compute_loss(train_out, targets, masks)[1] # box, obj, cls
# NMS
targets[:, 2:] *= torch.tensor((width, height, width, height), device=device) # to pixels
lb = [targets[targets[:, 0] == i, 1:] for i in range(nb)] if save_hybrid else [] # for autolabelling
with dt[2]:
preds = non_max_suppression(preds,
conf_thres,
iou_thres,
labels=lb,
multi_label=True,
agnostic=single_cls,
max_det=max_det,
nm=nm)
# Metrics
plot_masks = [] # masks for plotting
for si, (pred, proto) in enumerate(zip(preds, protos)):
labels = targets[targets[:, 0] == si, 1:]
nl, npr = labels.shape[0], pred.shape[0] # number of labels, predictions
path, shape = Path(paths[si]), shapes[si][0]
correct_masks = torch.zeros(npr, niou, dtype=torch.bool, device=device) # init
correct_bboxes = torch.zeros(npr, niou, dtype=torch.bool, device=device) # init
seen += 1
if npr == 0:
if nl:
stats.append((correct_masks, correct_bboxes, *torch.zeros((2, 0), device=device), labels[:, 0]))
if plots:
confusion_matrix.process_batch(detections=None, labels=labels[:, 0])
continue
# Masks
midx = [si] if overlap else targets[:, 0] == si
gt_masks = masks[midx]
pred_masks = process(proto, pred[:, 6:], pred[:, :4], shape=im[si].shape[1:])
# Predictions
if single_cls:
pred[:, 5] = 0
predn = pred.clone()
scale_boxes(im[si].shape[1:], predn[:, :4], shape, shapes[si][1]) # native-space pred
# Evaluate
if nl:
tbox = xywh2xyxy(labels[:, 1:5]) # target boxes
scale_boxes(im[si].shape[1:], tbox, shape, shapes[si][1]) # native-space labels
labelsn = torch.cat((labels[:, 0:1], tbox), 1) # native-space labels
correct_bboxes = process_batch(predn, labelsn, iouv)
correct_masks = process_batch(predn, labelsn, iouv, pred_masks, gt_masks, overlap=overlap, masks=True)
if plots:
confusion_matrix.process_batch(predn, labelsn)
stats.append((correct_masks, correct_bboxes, pred[:, 4], pred[:, 5], labels[:, 0])) # (conf, pcls, tcls)
pred_masks = torch.as_tensor(pred_masks, dtype=torch.uint8)
if plots and batch_i < 3:
plot_masks.append(pred_masks[:15].cpu()) # filter top 15 to plot
# Save/log
if save_txt:
save_one_txt(predn, save_conf, shape, file=save_dir / 'labels' / f'{path.stem}.txt')
if save_json:
pred_masks = scale_image(im[si].shape[1:],
pred_masks.permute(1, 2, 0).contiguous().cpu().numpy(), shape, shapes[si][1])
save_one_json(predn, jdict, path, class_map, pred_masks) # append to COCO-JSON dictionary
# callbacks.run('on_val_image_end', pred, predn, path, names, im[si])
# Plot images
if plots and batch_i < 3:
if len(plot_masks):
plot_masks = torch.cat(plot_masks, dim=0)
plot_images_and_masks(im, targets, masks, paths, save_dir / f'val_batch{batch_i}_labels.jpg', names)
plot_images_and_masks(im, output_to_target(preds, max_det=15), plot_masks, paths,
save_dir / f'val_batch{batch_i}_pred.jpg', names) # pred
# callbacks.run('on_val_batch_end')
# Compute metrics
stats = [torch.cat(x, 0).cpu().numpy() for x in zip(*stats)] # to numpy
if len(stats) and stats[0].any():
results = ap_per_class_box_and_mask(*stats, plot=plots, save_dir=save_dir, names=names)
metrics.update(results)
nt = np.bincount(stats[4].astype(int), minlength=nc) # number of targets per class
# Print results
pf = '%22s' + '%11i' * 2 + '%11.3g' * 8 # print format
LOGGER.info(pf % ("all", seen, nt.sum(), *metrics.mean_results()))
if nt.sum() == 0:
LOGGER.warning(f'WARNING ⚠️ no labels found in {task} set, can not compute metrics without labels')
# Print results per class
if (verbose or (nc < 50 and not training)) and nc > 1 and len(stats):
for i, c in enumerate(metrics.ap_class_index):
LOGGER.info(pf % (names[c], seen, nt[c], *metrics.class_result(i)))
# Print speeds
t = tuple(x.t / seen * 1E3 for x in dt) # speeds per image
if not training:
shape = (batch_size, 3, imgsz, imgsz)
LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {shape}' % t)
# Plots
if plots:
confusion_matrix.plot(save_dir=save_dir, names=list(names.values()))
# callbacks.run('on_val_end')
mp_bbox, mr_bbox, map50_bbox, map_bbox, mp_mask, mr_mask, map50_mask, map_mask = metrics.mean_results()
# Save JSON
if save_json and len(jdict):
w = Path(weights[0] if isinstance(weights, list) else weights).stem if weights is not None else '' # weights
anno_json = str(Path(data.get('path', '../coco')) / 'annotations/instances_val2017.json') # annotations json
pred_json = str(save_dir / f"{w}_predictions.json") # predictions json
LOGGER.info(f'\nEvaluating pycocotools mAP... saving {pred_json}...')
with open(pred_json, 'w') as f:
json.dump(jdict, f)
try: # https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval
anno = COCO(anno_json) # init annotations api
pred = anno.loadRes(pred_json) # init predictions api
results = []
for eval in COCOeval(anno, pred, 'bbox'), COCOeval(anno, pred, 'segm'):
if is_coco:
eval.params.imgIds = [int(Path(x).stem) for x in dataloader.dataset.im_files] # img ID to evaluate
eval.evaluate()
eval.accumulate()
eval.summarize()
results.extend(eval.stats[:2]) # update results (mAP@0.5:0.95, mAP@0.5)
map_bbox, map50_bbox, map_mask, map50_mask = results
except Exception as e:
LOGGER.info(f'pycocotools unable to run: {e}')
# Return results
model.float() # for training
if not training:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
final_metric = mp_bbox, mr_bbox, map50_bbox, map_bbox, mp_mask, mr_mask, map50_mask, map_mask
return (*final_metric, *(loss.cpu() / len(dataloader)).tolist()), metrics.get_maps(nc), t
def parse_opt():
parser = argparse.ArgumentParser()
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128-seg.yaml', help='dataset.yaml path')
parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolo-seg.pt', help='model path(s)')
parser.add_argument('--batch-size', type=int, default=32, help='batch size')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='inference size (pixels)')
parser.add_argument('--conf-thres', type=float, default=0.001, help='confidence threshold')
parser.add_argument('--iou-thres', type=float, default=0.6, help='NMS IoU threshold')
parser.add_argument('--max-det', type=int, default=300, help='maximum detections per image')
parser.add_argument('--task', default='val', help='train, val, test, speed or study')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
parser.add_argument('--single-cls', action='store_true', help='treat as single-class dataset')
parser.add_argument('--augment', action='store_true', help='augmented inference')
parser.add_argument('--verbose', action='store_true', help='report mAP by class')
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
parser.add_argument('--save-hybrid', action='store_true', help='save label+prediction hybrid results to *.txt')
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
parser.add_argument('--save-json', action='store_true', help='save a COCO-JSON results file')
parser.add_argument('--project', default=ROOT / 'runs/val-seg', help='save results to project/name')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')
opt = parser.parse_args()
opt.data = check_yaml(opt.data) # check YAML
# opt.save_json |= opt.data.endswith('coco.yaml')
opt.save_txt |= opt.save_hybrid
print_args(vars(opt))
return opt
def main(opt):
#check_requirements(requirements=ROOT / 'requirements.txt', exclude=('tensorboard', 'thop'))
if opt.task in ('train', 'val', 'test'): # run normally
if opt.conf_thres > 0.001: # https://github.com/ultralytics/yolov5/issues/1466
LOGGER.warning(f'WARNING ⚠️ confidence threshold {opt.conf_thres} > 0.001 produces invalid results')
if opt.save_hybrid:
LOGGER.warning('WARNING ⚠️ --save-hybrid returns high mAP from hybrid labels, not from predictions alone')
run(**vars(opt))
else:
weights = opt.weights if isinstance(opt.weights, list) else [opt.weights]
opt.half = torch.cuda.is_available() and opt.device != 'cpu' # FP16 for fastest results
if opt.task == 'speed': # speed benchmarks
# python val.py --task speed --data coco.yaml --batch 1 --weights yolo.pt...
opt.conf_thres, opt.iou_thres, opt.save_json = 0.25, 0.45, False
for opt.weights in weights:
run(**vars(opt), plots=False)
elif opt.task == 'study': # speed vs mAP benchmarks
# python val.py --task study --data coco.yaml --iou 0.7 --weights yolo.pt...
for opt.weights in weights:
f = f'study_{Path(opt.data).stem}_{Path(opt.weights).stem}.txt' # filename to save to
x, y = list(range(256, 1536 + 128, 128)), [] # x axis (image sizes), y axis
for opt.imgsz in x: # img-size
LOGGER.info(f'\nRunning {f} --imgsz {opt.imgsz}...')
r, _, t = run(**vars(opt), plots=False)
y.append(r + t) # results and times
np.savetxt(f, y, fmt='%10.4g') # save
os.system('zip -r study.zip study_*.txt')
plot_val_study(x=x) # plot
if __name__ == "__main__":
opt = parse_opt()
main(opt)
tools/reparameterization.ipynb 0 → 100644
View file @ f969ca34
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "4beac401",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"from models.yolo import Model"
]
},
{
"cell_type": "markdown",
"id": "8680f822",
"metadata": {},
"source": [
"## Convert YOLOv9-C"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "59f0198d",
"metadata": {},
"outputs": [],
"source": [
"device = torch.device(\"cpu\")\n",
"cfg = \"./models/detect/gelan-c.yaml\"\n",
"model = Model(cfg, ch=3, nc=80, anchors=3)\n",
"#model = model.half()\n",
"model = model.to(device)\n",
"_ = model.eval()\n",
"ckpt = torch.load('./yolov9-c.pt', map_location='cpu')\n",
"model.names = ckpt['model'].names\n",
"model.nc = ckpt['model'].nc"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2de7e1be",
"metadata": {},
"outputs": [],
"source": [
"idx = 0\n",
"for k, v in model.state_dict().items():\n",
" if \"model.{}.\".format(idx) in k:\n",
" if idx < 22:\n",
" kr = k.replace(\"model.{}.\".format(idx), \"model.{}.\".format(idx+1))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" elif \"model.{}.cv2.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.cv2.\".format(idx), \"model.{}.cv4.\".format(idx+16))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" elif \"model.{}.cv3.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.cv3.\".format(idx), \"model.{}.cv5.\".format(idx+16))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" elif \"model.{}.dfl.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.dfl.\".format(idx), \"model.{}.dfl2.\".format(idx+16))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" else:\n",
" while True:\n",
" idx += 1\n",
" if \"model.{}.\".format(idx) in k:\n",
" break\n",
" if idx < 22:\n",
" kr = k.replace(\"model.{}.\".format(idx), \"model.{}.\".format(idx+1))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" elif \"model.{}.cv2.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.cv2.\".format(idx), \"model.{}.cv4.\".format(idx+16))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" elif \"model.{}.cv3.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.cv3.\".format(idx), \"model.{}.cv5.\".format(idx+16))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" elif \"model.{}.dfl.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.dfl.\".format(idx), \"model.{}.dfl2.\".format(idx+16))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
"_ = model.eval()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "960796e3",
"metadata": {},
"outputs": [],
"source": [
"m_ckpt = {'model': model.half(),\n",
" 'optimizer': None,\n",
" 'best_fitness': None,\n",
" 'ema': None,\n",
" 'updates': None,\n",
" 'opt': None,\n",
" 'git': None,\n",
" 'date': None,\n",
" 'epoch': -1}\n",
"torch.save(m_ckpt, \"./yolov9-c-converted.pt\")"
]
},
{
"cell_type": "markdown",
"id": "47c6e6ae",
"metadata": {},
"source": [
"## Convert YOLOv9-E"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "801a1b7c",
"metadata": {},
"outputs": [],
"source": [
"device = torch.device(\"cpu\")\n",
"cfg = \"./models/detect/gelan-e.yaml\"\n",
"model = Model(cfg, ch=3, nc=80, anchors=3)\n",
"#model = model.half()\n",
"model = model.to(device)\n",
"_ = model.eval()\n",
"ckpt = torch.load('./yolov9-e.pt', map_location='cpu')\n",
"model.names = ckpt['model'].names\n",
"model.nc = ckpt['model'].nc"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a2ef4fe6",
"metadata": {},
"outputs": [],
"source": [
"idx = 0\n",
"for k, v in model.state_dict().items():\n",
" if \"model.{}.\".format(idx) in k:\n",
" if idx < 29:\n",
" kr = k.replace(\"model.{}.\".format(idx), \"model.{}.\".format(idx))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
" elif idx < 42:\n",
" kr = k.replace(\"model.{}.\".format(idx), \"model.{}.\".format(idx+7))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
" elif \"model.{}.cv2.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.cv2.\".format(idx), \"model.{}.cv4.\".format(idx+7))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
" elif \"model.{}.cv3.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.cv3.\".format(idx), \"model.{}.cv5.\".format(idx+7))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
" elif \"model.{}.dfl.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.dfl.\".format(idx), \"model.{}.dfl2.\".format(idx+7))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
" else:\n",
" while True:\n",
" idx += 1\n",
" if \"model.{}.\".format(idx) in k:\n",
" break\n",
" if idx < 29:\n",
" kr = k.replace(\"model.{}.\".format(idx), \"model.{}.\".format(idx))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
" elif idx < 42:\n",
" kr = k.replace(\"model.{}.\".format(idx), \"model.{}.\".format(idx+7))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
" elif \"model.{}.cv2.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.cv2.\".format(idx), \"model.{}.cv4.\".format(idx+7))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
" elif \"model.{}.cv3.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.cv3.\".format(idx), \"model.{}.cv5.\".format(idx+7))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
" elif \"model.{}.dfl.\".format(idx) in k:\n",
" kr = k.replace(\"model.{}.dfl.\".format(idx), \"model.{}.dfl2.\".format(idx+7))\n",
" model.state_dict()[k] -= model.state_dict()[k]\n",
" model.state_dict()[k] += ckpt['model'].state_dict()[kr]\n",
" print(k, \"perfectly matched!!\")\n",
"_ = model.eval()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "27bc1869",
"metadata": {},
"outputs": [],
"source": [
"m_ckpt = {'model': model.half(),\n",
" 'optimizer': None,\n",
" 'best_fitness': None,\n",
" 'ema': None,\n",
" 'updates': None,\n",
" 'opt': None,\n",
" 'git': None,\n",
" 'date': None,\n",
" 'epoch': -1}\n",
"torch.save(m_ckpt, \"./yolov9-e-converted.pt\")"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.12"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
train.py 0 → 100644
View file @ f969ca34
import argparse
import math
import os
import random
import sys
import time
from copy import deepcopy
from datetime import datetime
from pathlib import Path
import numpy as np
import torch
import torch.distributed as dist
import torch.nn as nn
import yaml
from torch.optim import lr_scheduler
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[0] # root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
import val as validate # for end-of-epoch mAP
from models.experimental import attempt_load
from models.yolo import Model
from utils.autoanchor import check_anchors
from utils.autobatch import check_train_batch_size
from utils.callbacks import Callbacks
from utils.dataloaders import create_dataloader
from utils.downloads import attempt_download, is_url
from utils.general import (LOGGER, TQDM_BAR_FORMAT, check_amp, check_dataset, check_file, check_img_size,
check_suffix, check_yaml, colorstr, get_latest_run, increment_path, init_seeds,
intersect_dicts, labels_to_class_weights, labels_to_image_weights, methods,
one_cycle, one_flat_cycle, print_args, print_mutation, strip_optimizer, yaml_save)
from utils.loggers import Loggers
from utils.loggers.comet.comet_utils import check_comet_resume
from utils.loss_tal import ComputeLoss
from utils.metrics import fitness
from utils.plots import plot_evolve
from utils.torch_utils import (EarlyStopping, ModelEMA, de_parallel, select_device, smart_DDP,
smart_optimizer, smart_resume, torch_distributed_zero_first)
LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1)) # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
GIT_INFO = None
def train(hyp, opt, device, callbacks): # hyp is path/to/hyp.yaml or hyp dictionary
save_dir, epochs, batch_size, weights, single_cls, evolve, data, cfg, resume, noval, nosave, workers, freeze = \
Path(opt.save_dir), opt.epochs, opt.batch_size, opt.weights, opt.single_cls, opt.evolve, opt.data, opt.cfg, \
opt.resume, opt.noval, opt.nosave, opt.workers, opt.freeze
callbacks.run('on_pretrain_routine_start')
# Directories
w = save_dir / 'weights' # weights dir
(w.parent if evolve else w).mkdir(parents=True, exist_ok=True) # make dir
last, best = w / 'last.pt', w / 'best.pt'
last_striped, best_striped = w / 'last_striped.pt', w / 'best_striped.pt'
# Hyperparameters
if isinstance(hyp, str):
with open(hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
LOGGER.info(colorstr('hyperparameters: ') + ', '.join(f'{k}={v}' for k, v in hyp.items()))
hyp['anchor_t'] = 5.0
opt.hyp = hyp.copy() # for saving hyps to checkpoints
# Save run settings
if not evolve:
yaml_save(save_dir / 'hyp.yaml', hyp)
yaml_save(save_dir / 'opt.yaml', vars(opt))
# Loggers
data_dict = None
if RANK in {-1, 0}:
loggers = Loggers(save_dir, weights, opt, hyp, LOGGER) # loggers instance
# Register actions
for k in methods(loggers):
callbacks.register_action(k, callback=getattr(loggers, k))
# Process custom dataset artifact link
data_dict = loggers.remote_dataset
if resume: # If resuming runs from remote artifact
weights, epochs, hyp, batch_size = opt.weights, opt.epochs, opt.hyp, opt.batch_size
# Config
plots = not evolve and not opt.noplots # create plots
cuda = device.type != 'cpu'
init_seeds(opt.seed + 1 + RANK, deterministic=True)
with torch_distributed_zero_first(LOCAL_RANK):
data_dict = data_dict or check_dataset(data) # check if None
train_path, val_path = data_dict['train'], data_dict['val']
nc = 1 if single_cls else int(data_dict['nc']) # number of classes
names = {0: 'item'} if single_cls and len(data_dict['names']) != 1 else data_dict['names'] # class names
#is_coco = isinstance(val_path, str) and val_path.endswith('coco/val2017.txt') # COCO dataset
is_coco = isinstance(val_path, str) and val_path.endswith('val2017.txt') # COCO dataset
# Model
check_suffix(weights, '.pt') # check weights
pretrained = weights.endswith('.pt')
if pretrained:
with torch_distributed_zero_first(LOCAL_RANK):
weights = attempt_download(weights) # download if not found locally
ckpt = torch.load(weights, map_location='cpu') # load checkpoint to CPU to avoid CUDA memory leak
model = Model(cfg or ckpt['model'].yaml, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # create
exclude = ['anchor'] if (cfg or hyp.get('anchors')) and not resume else [] # exclude keys
csd = ckpt['model'].float().state_dict() # checkpoint state_dict as FP32
csd = intersect_dicts(csd, model.state_dict(), exclude=exclude) # intersect
model.load_state_dict(csd, strict=False) # load
LOGGER.info(f'Transferred {len(csd)}/{len(model.state_dict())} items from {weights}') # report
else:
model = Model(cfg, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # create
amp = check_amp(model) # check AMP
# Freeze
freeze = [f'model.{x}.' for x in (freeze if len(freeze) > 1 else range(freeze[0]))] # layers to freeze
for k, v in model.named_parameters():
# v.requires_grad = True # train all layers TODO: uncomment this line as in master
# v.register_hook(lambda x: torch.nan_to_num(x)) # NaN to 0 (commented for erratic training results)
if any(x in k for x in freeze):
LOGGER.info(f'freezing {k}')
v.requires_grad = False
# Image size
gs = max(int(model.stride.max()), 32) # grid size (max stride)
imgsz = check_img_size(opt.imgsz, gs, floor=gs * 2) # verify imgsz is gs-multiple
# Batch size
if RANK == -1 and batch_size == -1: # single-GPU only, estimate best batch size
batch_size = check_train_batch_size(model, imgsz, amp)
loggers.on_params_update({"batch_size": batch_size})
# Optimizer
nbs = 64 # nominal batch size
accumulate = max(round(nbs / batch_size), 1) # accumulate loss before optimizing
hyp['weight_decay'] *= batch_size * accumulate / nbs # scale weight_decay
optimizer = smart_optimizer(model, opt.optimizer, hyp['lr0'], hyp['momentum'], hyp['weight_decay'])
# Scheduler
if opt.cos_lr:
lf = one_cycle(1, hyp['lrf'], epochs) # cosine 1->hyp['lrf']
elif opt.flat_cos_lr:
lf = one_flat_cycle(1, hyp['lrf'], epochs) # flat cosine 1->hyp['lrf']
elif opt.fixed_lr:
lf = lambda x: 1.0
else:
lf = lambda x: (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf'] # linear
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf)
# from utils.plots import plot_lr_scheduler; plot_lr_scheduler(optimizer, scheduler, epochs)
# EMA
ema = ModelEMA(model) if RANK in {-1, 0} else None
# Resume
best_fitness, start_epoch = 0.0, 0
if pretrained:
if resume:
best_fitness, start_epoch, epochs = smart_resume(ckpt, optimizer, ema, weights, epochs, resume)
del ckpt, csd
# DP mode
if cuda and RANK == -1 and torch.cuda.device_count() > 1:
LOGGER.warning('WARNING ⚠️ DP not recommended, use torch.distributed.run for best DDP Multi-GPU results.')
model = torch.nn.DataParallel(model)
# SyncBatchNorm
if opt.sync_bn and cuda and RANK != -1:
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
LOGGER.info('Using SyncBatchNorm()')
# Trainloader
train_loader, dataset = create_dataloader(train_path,
imgsz,
batch_size // WORLD_SIZE,
gs,
single_cls,
hyp=hyp,
augment=True,
cache=None if opt.cache == 'val' else opt.cache,
rect=opt.rect,
rank=LOCAL_RANK,
workers=workers,
image_weights=opt.image_weights,
close_mosaic=opt.close_mosaic != 0,
quad=opt.quad,
prefix=colorstr('train: '),
shuffle=True,
min_items=opt.min_items)
labels = np.concatenate(dataset.labels, 0)
mlc = int(labels[:, 0].max()) # max label class
assert mlc < nc, f'Label class {mlc} exceeds nc={nc} in {data}. Possible class labels are 0-{nc - 1}'
# Process 0
if RANK in {-1, 0}:
val_loader = create_dataloader(val_path,
imgsz,
batch_size // WORLD_SIZE * 2,
gs,
single_cls,
hyp=hyp,
cache=None if noval else opt.cache,
rect=True,
rank=-1,
workers=workers * 2,
pad=0.5,
prefix=colorstr('val: '))[0]
if not resume:
# if not opt.noautoanchor:
# check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz) # run AutoAnchor
model.half().float() # pre-reduce anchor precision
callbacks.run('on_pretrain_routine_end', labels, names)
# DDP mode
if cuda and RANK != -1:
model = smart_DDP(model)
# Model attributes
nl = de_parallel(model).model[-1].nl # number of detection layers (to scale hyps)
#hyp['box'] *= 3 / nl # scale to layers
#hyp['cls'] *= nc / 80 * 3 / nl # scale to classes and layers
#hyp['obj'] *= (imgsz / 640) ** 2 * 3 / nl # scale to image size and layers
hyp['label_smoothing'] = opt.label_smoothing
model.nc = nc # attach number of classes to model
model.hyp = hyp # attach hyperparameters to model
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) * nc # attach class weights
model.names = names
# Start training
t0 = time.time()
nb = len(train_loader) # number of batches
nw = max(round(hyp['warmup_epochs'] * nb), 100) # number of warmup iterations, max(3 epochs, 100 iterations)
# nw = min(nw, (epochs - start_epoch) / 2 * nb) # limit warmup to < 1/2 of training
last_opt_step = -1
maps = np.zeros(nc) # mAP per class
results = (0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
scheduler.last_epoch = start_epoch - 1 # do not move
scaler = torch.cuda.amp.GradScaler(enabled=amp)
stopper, stop = EarlyStopping(patience=opt.patience), False
compute_loss = ComputeLoss(model) # init loss class
callbacks.run('on_train_start')
LOGGER.info(f'Image sizes {imgsz} train, {imgsz} val\n'
f'Using {train_loader.num_workers * WORLD_SIZE} dataloader workers\n'
f"Logging results to {colorstr('bold', save_dir)}\n"
f'Starting training for {epochs} epochs...')
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
callbacks.run('on_train_epoch_start')
model.train()
# Update image weights (optional, single-GPU only)
if opt.image_weights:
cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 / nc # class weights
iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw) # image weights
dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n) # rand weighted idx
if epoch == (epochs - opt.close_mosaic):
LOGGER.info("Closing dataloader mosaic")
dataset.mosaic = False
# Update mosaic border (optional)
# b = int(random.uniform(0.25 * imgsz, 0.75 * imgsz + gs) // gs * gs)
# dataset.mosaic_border = [b - imgsz, -b] # height, width borders
mloss = torch.zeros(3, device=device) # mean losses
if RANK != -1:
train_loader.sampler.set_epoch(epoch)
pbar = enumerate(train_loader)
LOGGER.info(('\n' + '%11s' * 7) % ('Epoch', 'GPU_mem', 'box_loss', 'cls_loss', 'dfl_loss', 'Instances', 'Size'))
if RANK in {-1, 0}:
pbar = tqdm(pbar, total=nb, bar_format=TQDM_BAR_FORMAT) # progress bar
optimizer.zero_grad()
for i, (imgs, targets, paths, _) in pbar: # batch -------------------------------------------------------------
callbacks.run('on_train_batch_start')
ni = i + nb * epoch # number integrated batches (since train start)
imgs = imgs.to(device, non_blocking=True).float() / 255 # uint8 to float32, 0-255 to 0.0-1.0
# Warmup
if ni <= nw:
xi = [0, nw] # x interp
# compute_loss.gr = np.interp(ni, xi, [0.0, 1.0]) # iou loss ratio (obj_loss = 1.0 or iou)
accumulate = max(1, np.interp(ni, xi, [1, nbs / batch_size]).round())
for j, x in enumerate(optimizer.param_groups):
# bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
x['lr'] = np.interp(ni, xi, [hyp['warmup_bias_lr'] if j == 0 else 0.0, x['initial_lr'] * lf(epoch)])
if 'momentum' in x:
x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])
# Multi-scale
if opt.multi_scale:
sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs # size
sf = sz / max(imgs.shape[2:]) # scale factor
if sf != 1:
ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]] # new shape (stretched to gs-multiple)
imgs = nn.functional.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)
# Forward
with torch.cuda.amp.autocast(amp):
pred = model(imgs) # forward
loss, loss_items = compute_loss(pred, targets.to(device)) # loss scaled by batch_size
if RANK != -1:
loss *= WORLD_SIZE # gradient averaged between devices in DDP mode
if opt.quad:
loss *= 4.
# Backward
scaler.scale(loss).backward()
# Optimize - https://pytorch.org/docs/master/notes/amp_examples.html
if ni - last_opt_step >= accumulate:
scaler.unscale_(optimizer) # unscale gradients
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=10.0) # clip gradients
scaler.step(optimizer) # optimizer.step
scaler.update()
optimizer.zero_grad()
if ema:
ema.update(model)
last_opt_step = ni
# Log
if RANK in {-1, 0}:
mloss = (mloss * i + loss_items) / (i + 1) # update mean losses
mem = f'{torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0:.3g}G' # (GB)
pbar.set_description(('%11s' * 2 + '%11.4g' * 5) %
(f'{epoch}/{epochs - 1}', mem, *mloss, targets.shape[0], imgs.shape[-1]))
callbacks.run('on_train_batch_end', model, ni, imgs, targets, paths, list(mloss))
if callbacks.stop_training:
return
# end batch ------------------------------------------------------------------------------------------------
# Scheduler
lr = [x['lr'] for x in optimizer.param_groups] # for loggers
scheduler.step()
if RANK in {-1, 0}:
# mAP
callbacks.run('on_train_epoch_end', epoch=epoch)
ema.update_attr(model, include=['yaml', 'nc', 'hyp', 'names', 'stride', 'class_weights'])
final_epoch = (epoch + 1 == epochs) or stopper.possible_stop
if not noval or final_epoch: # Calculate mAP
results, maps, _ = validate.run(data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
half=amp,
model=ema.ema,
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
plots=False,
callbacks=callbacks,
compute_loss=compute_loss)
# Update best mAP
fi = fitness(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
stop = stopper(epoch=epoch, fitness=fi) # early stop check
if fi > best_fitness:
best_fitness = fi
log_vals = list(mloss) + list(results) + lr
callbacks.run('on_fit_epoch_end', log_vals, epoch, best_fitness, fi)
# Save model
if (not nosave) or (final_epoch and not evolve): # if save
ckpt = {
'epoch': epoch,
'best_fitness': best_fitness,
'model': deepcopy(de_parallel(model)).half(),
'ema': deepcopy(ema.ema).half(),
'updates': ema.updates,
'optimizer': optimizer.state_dict(),
'opt': vars(opt),
'git': GIT_INFO, # {remote, branch, commit} if a git repo
'date': datetime.now().isoformat()}
# Save last, best and delete
torch.save(ckpt, last)
if best_fitness == fi:
torch.save(ckpt, best)
if opt.save_period > 0 and epoch % opt.save_period == 0:
torch.save(ckpt, w / f'epoch{epoch}.pt')
del ckpt
callbacks.run('on_model_save', last, epoch, final_epoch, best_fitness, fi)
# EarlyStopping
if RANK != -1: # if DDP training
broadcast_list = [stop if RANK == 0 else None]
dist.broadcast_object_list(broadcast_list, 0) # broadcast 'stop' to all ranks
if RANK != 0:
stop = broadcast_list[0]
if stop:
break # must break all DDP ranks
# end epoch ----------------------------------------------------------------------------------------------------
# end training -----------------------------------------------------------------------------------------------------
if RANK in {-1, 0}:
LOGGER.info(f'\n{epoch - start_epoch + 1} epochs completed in {(time.time() - t0) / 3600:.3f} hours.')
for f in last, best:
if f.exists():
if f is last:
strip_optimizer(f, last_striped) # strip optimizers
else:
strip_optimizer(f, best_striped) # strip optimizers
if f is best:
LOGGER.info(f'\nValidating {f}...')
results, _, _ = validate.run(
data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
model=attempt_load(f, device).half(),
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
save_json=is_coco,
verbose=True,
plots=plots,
callbacks=callbacks,
compute_loss=compute_loss) # val best model with plots
if is_coco:
callbacks.run('on_fit_epoch_end', list(mloss) + list(results) + lr, epoch, best_fitness, fi)
callbacks.run('on_train_end', last, best, epoch, results)
torch.cuda.empty_cache()
return results
def parse_opt(known=False):
parser = argparse.ArgumentParser()
# parser.add_argument('--weights', type=str, default=ROOT / 'yolo.pt', help='initial weights path')
# parser.add_argument('--cfg', type=str, default='', help='model.yaml path')
parser.add_argument('--weights', type=str, default='', help='initial weights path')
parser.add_argument('--cfg', type=str, default='yolo.yaml', help='model.yaml path')
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='dataset.yaml path')
parser.add_argument('--hyp', type=str, default=ROOT / 'data/hyps/hyp.scratch-low.yaml', help='hyperparameters path')
parser.add_argument('--epochs', type=int, default=100, help='total training epochs')
parser.add_argument('--batch-size', type=int, default=16, help='total batch size for all GPUs, -1 for autobatch')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='train, val image size (pixels)')
parser.add_argument('--rect', action='store_true', help='rectangular training')
parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
parser.add_argument('--noval', action='store_true', help='only validate final epoch')
parser.add_argument('--noautoanchor', action='store_true', help='disable AutoAnchor')
parser.add_argument('--noplots', action='store_true', help='save no plot files')
parser.add_argument('--evolve', type=int, nargs='?', const=300, help='evolve hyperparameters for x generations')
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
parser.add_argument('--cache', type=str, nargs='?', const='ram', help='image --cache ram/disk')
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
parser.add_argument('--single-cls', action='store_true', help='train multi-class data as single-class')
parser.add_argument('--optimizer', type=str, choices=['SGD', 'Adam', 'AdamW', 'LION'], default='SGD', help='optimizer')
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
parser.add_argument('--project', default=ROOT / 'runs/train', help='save to project/name')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--quad', action='store_true', help='quad dataloader')
parser.add_argument('--cos-lr', action='store_true', help='cosine LR scheduler')
parser.add_argument('--flat-cos-lr', action='store_true', help='flat cosine LR scheduler')
parser.add_argument('--fixed-lr', action='store_true', help='fixed LR scheduler')
parser.add_argument('--label-smoothing', type=float, default=0.0, help='Label smoothing epsilon')
parser.add_argument('--patience', type=int, default=100, help='EarlyStopping patience (epochs without improvement)')
parser.add_argument('--freeze', nargs='+', type=int, default=[0], help='Freeze layers: backbone=10, first3=0 1 2')
parser.add_argument('--save-period', type=int, default=-1, help='Save checkpoint every x epochs (disabled if < 1)')
parser.add_argument('--seed', type=int, default=0, help='Global training seed')
parser.add_argument('--local_rank', type=int, default=-1, help='Automatic DDP Multi-GPU argument, do not modify')
parser.add_argument('--min-items', type=int, default=0, help='Experimental')
parser.add_argument('--close-mosaic', type=int, default=0, help='Experimental')
# Logger arguments
parser.add_argument('--entity', default=None, help='Entity')
parser.add_argument('--upload_dataset', nargs='?', const=True, default=False, help='Upload data, "val" option')
parser.add_argument('--bbox_interval', type=int, default=-1, help='Set bounding-box image logging interval')
parser.add_argument('--artifact_alias', type=str, default='latest', help='Version of dataset artifact to use')
return parser.parse_known_args()[0] if known else parser.parse_args()
def main(opt, callbacks=Callbacks()):
# Checks
if RANK in {-1, 0}:
print_args(vars(opt))
# Resume (from specified or most recent last.pt)
if opt.resume and not check_comet_resume(opt) and not opt.evolve:
last = Path(check_file(opt.resume) if isinstance(opt.resume, str) else get_latest_run())
opt_yaml = last.parent.parent / 'opt.yaml' # train options yaml
opt_data = opt.data # original dataset
if opt_yaml.is_file():
with open(opt_yaml, errors='ignore') as f:
d = yaml.safe_load(f)
else:
d = torch.load(last, map_location='cpu')['opt']
opt = argparse.Namespace(**d) # replace
opt.cfg, opt.weights, opt.resume = '', str(last), True # reinstate
if is_url(opt_data):
opt.data = check_file(opt_data) # avoid HUB resume auth timeout
else:
opt.data, opt.cfg, opt.hyp, opt.weights, opt.project = \
check_file(opt.data), check_yaml(opt.cfg), check_yaml(opt.hyp), str(opt.weights), str(opt.project) # checks
assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
if opt.evolve:
if opt.project == str(ROOT / 'runs/train'): # if default project name, rename to runs/evolve
opt.project = str(ROOT / 'runs/evolve')
opt.exist_ok, opt.resume = opt.resume, False # pass resume to exist_ok and disable resume
if opt.name == 'cfg':
opt.name = Path(opt.cfg).stem # use model.yaml as name
opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok))
# DDP mode
device = select_device(opt.device, batch_size=opt.batch_size)
if LOCAL_RANK != -1:
msg = 'is not compatible with YOLO Multi-GPU DDP training'
assert not opt.image_weights, f'--image-weights {msg}'
assert not opt.evolve, f'--evolve {msg}'
assert opt.batch_size != -1, f'AutoBatch with --batch-size -1 {msg}, please pass a valid --batch-size'
assert opt.batch_size % WORLD_SIZE == 0, f'--batch-size {opt.batch_size} must be multiple of WORLD_SIZE'
assert torch.cuda.device_count() > LOCAL_RANK, 'insufficient CUDA devices for DDP command'
torch.cuda.set_device(LOCAL_RANK)
device = torch.device('cuda', LOCAL_RANK)
dist.init_process_group(backend="nccl" if dist.is_nccl_available() else "gloo")
# Train
if not opt.evolve:
train(opt.hyp, opt, device, callbacks)
# Evolve hyperparameters (optional)
else:
# Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
meta = {
'lr0': (1, 1e-5, 1e-1), # initial learning rate (SGD=1E-2, Adam=1E-3)
'lrf': (1, 0.01, 1.0), # final OneCycleLR learning rate (lr0 * lrf)
'momentum': (0.3, 0.6, 0.98), # SGD momentum/Adam beta1
'weight_decay': (1, 0.0, 0.001), # optimizer weight decay
'warmup_epochs': (1, 0.0, 5.0), # warmup epochs (fractions ok)
'warmup_momentum': (1, 0.0, 0.95), # warmup initial momentum
'warmup_bias_lr': (1, 0.0, 0.2), # warmup initial bias lr
'box': (1, 0.02, 0.2), # box loss gain
'cls': (1, 0.2, 4.0), # cls loss gain
'cls_pw': (1, 0.5, 2.0), # cls BCELoss positive_weight
'obj': (1, 0.2, 4.0), # obj loss gain (scale with pixels)
'obj_pw': (1, 0.5, 2.0), # obj BCELoss positive_weight
'iou_t': (0, 0.1, 0.7), # IoU training threshold
'anchor_t': (1, 2.0, 8.0), # anchor-multiple threshold
'anchors': (2, 2.0, 10.0), # anchors per output grid (0 to ignore)
'fl_gamma': (0, 0.0, 2.0), # focal loss gamma (efficientDet default gamma=1.5)
'hsv_h': (1, 0.0, 0.1), # image HSV-Hue augmentation (fraction)
'hsv_s': (1, 0.0, 0.9), # image HSV-Saturation augmentation (fraction)
'hsv_v': (1, 0.0, 0.9), # image HSV-Value augmentation (fraction)
'degrees': (1, 0.0, 45.0), # image rotation (+/- deg)
'translate': (1, 0.0, 0.9), # image translation (+/- fraction)
'scale': (1, 0.0, 0.9), # image scale (+/- gain)
'shear': (1, 0.0, 10.0), # image shear (+/- deg)
'perspective': (0, 0.0, 0.001), # image perspective (+/- fraction), range 0-0.001
'flipud': (1, 0.0, 1.0), # image flip up-down (probability)
'fliplr': (0, 0.0, 1.0), # image flip left-right (probability)
'mosaic': (1, 0.0, 1.0), # image mixup (probability)
'mixup': (1, 0.0, 1.0), # image mixup (probability)
'copy_paste': (1, 0.0, 1.0)} # segment copy-paste (probability)
with open(opt.hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
if 'anchors' not in hyp: # anchors commented in hyp.yaml
hyp['anchors'] = 3
if opt.noautoanchor:
del hyp['anchors'], meta['anchors']
opt.noval, opt.nosave, save_dir = True, True, Path(opt.save_dir) # only val/save final epoch
# ei = [isinstance(x, (int, float)) for x in hyp.values()] # evolvable indices
evolve_yaml, evolve_csv = save_dir / 'hyp_evolve.yaml', save_dir / 'evolve.csv'
if opt.bucket:
os.system(f'gsutil cp gs://{opt.bucket}/evolve.csv {evolve_csv}') # download evolve.csv if exists
for _ in range(opt.evolve): # generations to evolve
if evolve_csv.exists(): # if evolve.csv exists: select best hyps and mutate
# Select parent(s)
parent = 'single' # parent selection method: 'single' or 'weighted'
x = np.loadtxt(evolve_csv, ndmin=2, delimiter=',', skiprows=1)
n = min(5, len(x)) # number of previous results to consider
x = x[np.argsort(-fitness(x))][:n] # top n mutations
w = fitness(x) - fitness(x).min() + 1E-6 # weights (sum > 0)
if parent == 'single' or len(x) == 1:
# x = x[random.randint(0, n - 1)] # random selection
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
elif parent == 'weighted':
x = (x * w.reshape(n, 1)).sum(0) / w.sum() # weighted combination
# Mutate
mp, s = 0.8, 0.2 # mutation probability, sigma
npr = np.random
npr.seed(int(time.time()))
g = np.array([meta[k][0] for k in hyp.keys()]) # gains 0-1
ng = len(meta)
v = np.ones(ng)
while all(v == 1): # mutate until a change occurs (prevent duplicates)
v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
for i, k in enumerate(hyp.keys()): # plt.hist(v.ravel(), 300)
hyp[k] = float(x[i + 7] * v[i]) # mutate
# Constrain to limits
for k, v in meta.items():
hyp[k] = max(hyp[k], v[1]) # lower limit
hyp[k] = min(hyp[k], v[2]) # upper limit
hyp[k] = round(hyp[k], 5) # significant digits
# Train mutation
results = train(hyp.copy(), opt, device, callbacks)
callbacks = Callbacks()
# Write mutation results
keys = ('metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95', 'val/box_loss',
'val/obj_loss', 'val/cls_loss')
print_mutation(keys, results, hyp.copy(), save_dir, opt.bucket)
# Plot results
plot_evolve(evolve_csv)
LOGGER.info(f'Hyperparameter evolution finished {opt.evolve} generations\n'
f"Results saved to {colorstr('bold', save_dir)}\n"
f'Usage example: $ python train.py --hyp {evolve_yaml}')
def run(**kwargs):
# Usage: import train; train.run(data='coco128.yaml', imgsz=320, weights='yolo.pt')
opt = parse_opt(True)
for k, v in kwargs.items():
setattr(opt, k, v)
main(opt)
return opt
if __name__ == "__main__":
opt = parse_opt()
main(opt)
train_dual.py 0 → 100644
View file @ f969ca34
import argparse
import math
import os
import random
import sys
import time
from copy import deepcopy
from datetime import datetime
from pathlib import Path
import numpy as np
import torch
import torch.distributed as dist
import torch.nn as nn
import yaml
from torch.optim import lr_scheduler
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[0] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
import val_dual as validate # for end-of-epoch mAP
from models.experimental import attempt_load
from models.yolo import Model
from utils.autoanchor import check_anchors
from utils.autobatch import check_train_batch_size
from utils.callbacks import Callbacks
from utils.dataloaders import create_dataloader
from utils.downloads import attempt_download, is_url
from utils.general import (LOGGER, TQDM_BAR_FORMAT, check_amp, check_dataset, check_file, check_git_info,
check_git_status, check_img_size, check_requirements, check_suffix, check_yaml, colorstr,
get_latest_run, increment_path, init_seeds, intersect_dicts, labels_to_class_weights,
labels_to_image_weights, methods, one_cycle, print_args, print_mutation, strip_optimizer,
yaml_save, one_flat_cycle)
from utils.loggers import Loggers
from utils.loggers.comet.comet_utils import check_comet_resume
from utils.loss_tal_dual import ComputeLoss
#from utils.loss_tal_dual import ComputeLossLH as ComputeLoss
#from utils.loss_tal_dual import ComputeLossLHCF as ComputeLoss
from utils.metrics import fitness
from utils.plots import plot_evolve
from utils.torch_utils import (EarlyStopping, ModelEMA, de_parallel, select_device, smart_DDP, smart_optimizer,
smart_resume, torch_distributed_zero_first)
LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1)) # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
GIT_INFO = None#check_git_info()
def train(hyp, opt, device, callbacks): # hyp is path/to/hyp.yaml or hyp dictionary
save_dir, epochs, batch_size, weights, single_cls, evolve, data, cfg, resume, noval, nosave, workers, freeze = \
Path(opt.save_dir), opt.epochs, opt.batch_size, opt.weights, opt.single_cls, opt.evolve, opt.data, opt.cfg, \
opt.resume, opt.noval, opt.nosave, opt.workers, opt.freeze
callbacks.run('on_pretrain_routine_start')
# Directories
w = save_dir / 'weights' # weights dir
(w.parent if evolve else w).mkdir(parents=True, exist_ok=True) # make dir
last, best = w / 'last.pt', w / 'best.pt'
# Hyperparameters
if isinstance(hyp, str):
with open(hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
LOGGER.info(colorstr('hyperparameters: ') + ', '.join(f'{k}={v}' for k, v in hyp.items()))
hyp['anchor_t'] = 5.0
opt.hyp = hyp.copy() # for saving hyps to checkpoints
# Save run settings
if not evolve:
yaml_save(save_dir / 'hyp.yaml', hyp)
yaml_save(save_dir / 'opt.yaml', vars(opt))
# Loggers
data_dict = None
if RANK in {-1, 0}:
loggers = Loggers(save_dir, weights, opt, hyp, LOGGER) # loggers instance
# Register actions
for k in methods(loggers):
callbacks.register_action(k, callback=getattr(loggers, k))
# Process custom dataset artifact link
data_dict = loggers.remote_dataset
if resume: # If resuming runs from remote artifact
weights, epochs, hyp, batch_size = opt.weights, opt.epochs, opt.hyp, opt.batch_size
# Config
plots = not evolve and not opt.noplots # create plots
cuda = device.type != 'cpu'
init_seeds(opt.seed + 1 + RANK, deterministic=True)
with torch_distributed_zero_first(LOCAL_RANK):
data_dict = data_dict or check_dataset(data) # check if None
train_path, val_path = data_dict['train'], data_dict['val']
nc = 1 if single_cls else int(data_dict['nc']) # number of classes
names = {0: 'item'} if single_cls and len(data_dict['names']) != 1 else data_dict['names'] # class names
#is_coco = isinstance(val_path, str) and val_path.endswith('coco/val2017.txt') # COCO dataset
is_coco = isinstance(val_path, str) and val_path.endswith('val2017.txt') # COCO dataset
# Model
check_suffix(weights, '.pt') # check weights
pretrained = weights.endswith('.pt')
if pretrained:
with torch_distributed_zero_first(LOCAL_RANK):
weights = attempt_download(weights) # download if not found locally
ckpt = torch.load(weights, map_location='cpu') # load checkpoint to CPU to avoid CUDA memory leak
model = Model(cfg or ckpt['model'].yaml, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # create
exclude = ['anchor'] if (cfg or hyp.get('anchors')) and not resume else [] # exclude keys
csd = ckpt['model'].float().state_dict() # checkpoint state_dict as FP32
csd = intersect_dicts(csd, model.state_dict(), exclude=exclude) # intersect
model.load_state_dict(csd, strict=False) # load
LOGGER.info(f'Transferred {len(csd)}/{len(model.state_dict())} items from {weights}') # report
else:
model = Model(cfg, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # create
amp = check_amp(model) # check AMP
# Freeze
freeze = [f'model.{x}.' for x in (freeze if len(freeze) > 1 else range(freeze[0]))] # layers to freeze
for k, v in model.named_parameters():
# v.requires_grad = True # train all layers TODO: uncomment this line as in master
# v.register_hook(lambda x: torch.nan_to_num(x)) # NaN to 0 (commented for erratic training results)
if any(x in k for x in freeze):
LOGGER.info(f'freezing {k}')
v.requires_grad = False
# Image size
gs = max(int(model.stride.max()), 32) # grid size (max stride)
imgsz = check_img_size(opt.imgsz, gs, floor=gs * 2) # verify imgsz is gs-multiple
# Batch size
if RANK == -1 and batch_size == -1: # single-GPU only, estimate best batch size
batch_size = check_train_batch_size(model, imgsz, amp)
loggers.on_params_update({"batch_size": batch_size})
# Optimizer
nbs = 64 # nominal batch size
accumulate = max(round(nbs / batch_size), 1) # accumulate loss before optimizing
hyp['weight_decay'] *= batch_size * accumulate / nbs # scale weight_decay
optimizer = smart_optimizer(model, opt.optimizer, hyp['lr0'], hyp['momentum'], hyp['weight_decay'])
# Scheduler
if opt.cos_lr:
lf = one_cycle(1, hyp['lrf'], epochs) # cosine 1->hyp['lrf']
elif opt.flat_cos_lr:
lf = one_flat_cycle(1, hyp['lrf'], epochs) # flat cosine 1->hyp['lrf']
elif opt.fixed_lr:
lf = lambda x: 1.0
else:
lf = lambda x: (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf'] # linear
# def lf(x): # saw
# return (1 - (x % 30) / 30) * (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf']
#
# def lf(x): # triangle start at min
# return 2 * abs(x / 30 - math.floor(x / 30 + 1 / 2)) * (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf']
#
# def lf(x): # triangle start at max
# return 2 * abs(x / 32 + .5 - math.floor(x / 32 + 1)) * (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf']
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf)
# from utils.plots import plot_lr_scheduler; plot_lr_scheduler(optimizer, scheduler, epochs)
# EMA
ema = ModelEMA(model) if RANK in {-1, 0} else None
# Resume
best_fitness, start_epoch = 0.0, 0
if pretrained:
if resume:
best_fitness, start_epoch, epochs = smart_resume(ckpt, optimizer, ema, weights, epochs, resume)
del ckpt, csd
# DP mode
if cuda and RANK == -1 and torch.cuda.device_count() > 1:
LOGGER.warning('WARNING ⚠️ DP not recommended, use torch.distributed.run for best DDP Multi-GPU results.')
model = torch.nn.DataParallel(model)
# SyncBatchNorm
if opt.sync_bn and cuda and RANK != -1:
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
LOGGER.info('Using SyncBatchNorm()')
# Trainloader
train_loader, dataset = create_dataloader(train_path,
imgsz,
batch_size // WORLD_SIZE,
gs,
single_cls,
hyp=hyp,
augment=True,
cache=None if opt.cache == 'val' else opt.cache,
rect=opt.rect,
rank=LOCAL_RANK,
workers=workers,
image_weights=opt.image_weights,
close_mosaic=opt.close_mosaic != 0,
quad=opt.quad,
prefix=colorstr('train: '),
shuffle=True,
min_items=opt.min_items)
labels = np.concatenate(dataset.labels, 0)
mlc = int(labels[:, 0].max()) # max label class
assert mlc < nc, f'Label class {mlc} exceeds nc={nc} in {data}. Possible class labels are 0-{nc - 1}'
# Process 0
if RANK in {-1, 0}:
val_loader = create_dataloader(val_path,
imgsz,
batch_size // WORLD_SIZE * 2,
gs,
single_cls,
hyp=hyp,
cache=None if noval else opt.cache,
rect=True,
rank=-1,
workers=workers * 2,
pad=0.5,
prefix=colorstr('val: '))[0]
if not resume:
# if not opt.noautoanchor:
# check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz) # run AutoAnchor
model.half().float() # pre-reduce anchor precision
callbacks.run('on_pretrain_routine_end', labels, names)
# DDP mode
if cuda and RANK != -1:
model = smart_DDP(model)
# Model attributes
nl = de_parallel(model).model[-1].nl # number of detection layers (to scale hyps)
#hyp['box'] *= 3 / nl # scale to layers
#hyp['cls'] *= nc / 80 * 3 / nl # scale to classes and layers
#hyp['obj'] *= (imgsz / 640) ** 2 * 3 / nl # scale to image size and layers
hyp['label_smoothing'] = opt.label_smoothing
model.nc = nc # attach number of classes to model
model.hyp = hyp # attach hyperparameters to model
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) * nc # attach class weights
model.names = names
# Start training
t0 = time.time()
nb = len(train_loader) # number of batches
nw = max(round(hyp['warmup_epochs'] * nb), 100) # number of warmup iterations, max(3 epochs, 100 iterations)
# nw = min(nw, (epochs - start_epoch) / 2 * nb) # limit warmup to < 1/2 of training
last_opt_step = -1
maps = np.zeros(nc) # mAP per class
results = (0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
scheduler.last_epoch = start_epoch - 1 # do not move
scaler = torch.cuda.amp.GradScaler(enabled=amp)
stopper, stop = EarlyStopping(patience=opt.patience), False
compute_loss = ComputeLoss(model) # init loss class
callbacks.run('on_train_start')
LOGGER.info(f'Image sizes {imgsz} train, {imgsz} val\n'
f'Using {train_loader.num_workers * WORLD_SIZE} dataloader workers\n'
f"Logging results to {colorstr('bold', save_dir)}\n"
f'Starting training for {epochs} epochs...')
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
callbacks.run('on_train_epoch_start')
model.train()
# Update image weights (optional, single-GPU only)
if opt.image_weights:
cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 / nc # class weights
iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw) # image weights
dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n) # rand weighted idx
if epoch == (epochs - opt.close_mosaic):
LOGGER.info("Closing dataloader mosaic")
dataset.mosaic = False
# Update mosaic border (optional)
# b = int(random.uniform(0.25 * imgsz, 0.75 * imgsz + gs) // gs * gs)
# dataset.mosaic_border = [b - imgsz, -b] # height, width borders
mloss = torch.zeros(3, device=device) # mean losses
if RANK != -1:
train_loader.sampler.set_epoch(epoch)
pbar = enumerate(train_loader)
LOGGER.info(('\n' + '%11s' * 7) % ('Epoch', 'GPU_mem', 'box_loss', 'cls_loss', 'dfl_loss', 'Instances', 'Size'))
if RANK in {-1, 0}:
pbar = tqdm(pbar, total=nb, bar_format=TQDM_BAR_FORMAT) # progress bar
optimizer.zero_grad()
for i, (imgs, targets, paths, _) in pbar: # batch -------------------------------------------------------------
callbacks.run('on_train_batch_start')
ni = i + nb * epoch # number integrated batches (since train start)
imgs = imgs.to(device, non_blocking=True).float() / 255 # uint8 to float32, 0-255 to 0.0-1.0
# Warmup
if ni <= nw:
xi = [0, nw] # x interp
# compute_loss.gr = np.interp(ni, xi, [0.0, 1.0]) # iou loss ratio (obj_loss = 1.0 or iou)
accumulate = max(1, np.interp(ni, xi, [1, nbs / batch_size]).round())
for j, x in enumerate(optimizer.param_groups):
# bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
x['lr'] = np.interp(ni, xi, [hyp['warmup_bias_lr'] if j == 0 else 0.0, x['initial_lr'] * lf(epoch)])
if 'momentum' in x:
x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])
# Multi-scale
if opt.multi_scale:
sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs # size
sf = sz / max(imgs.shape[2:]) # scale factor
if sf != 1:
ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]] # new shape (stretched to gs-multiple)
imgs = nn.functional.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)
# Forward
with torch.cuda.amp.autocast(amp):
pred = model(imgs) # forward
loss, loss_items = compute_loss(pred, targets.to(device)) # loss scaled by batch_size
if RANK != -1:
loss *= WORLD_SIZE # gradient averaged between devices in DDP mode
if opt.quad:
loss *= 4.
# Backward
scaler.scale(loss).backward()
# Optimize - https://pytorch.org/docs/master/notes/amp_examples.html
if ni - last_opt_step >= accumulate:
scaler.unscale_(optimizer) # unscale gradients
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=10.0) # clip gradients
scaler.step(optimizer) # optimizer.step
scaler.update()
optimizer.zero_grad()
if ema:
ema.update(model)
last_opt_step = ni
# Log
if RANK in {-1, 0}:
mloss = (mloss * i + loss_items) / (i + 1) # update mean losses
mem = f'{torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0:.3g}G' # (GB)
pbar.set_description(('%11s' * 2 + '%11.4g' * 5) %
(f'{epoch}/{epochs - 1}', mem, *mloss, targets.shape[0], imgs.shape[-1]))
callbacks.run('on_train_batch_end', model, ni, imgs, targets, paths, list(mloss))
if callbacks.stop_training:
return
# end batch ------------------------------------------------------------------------------------------------
# Scheduler
lr = [x['lr'] for x in optimizer.param_groups] # for loggers
scheduler.step()
if RANK in {-1, 0}:
# mAP
callbacks.run('on_train_epoch_end', epoch=epoch)
ema.update_attr(model, include=['yaml', 'nc', 'hyp', 'names', 'stride', 'class_weights'])
final_epoch = (epoch + 1 == epochs) or stopper.possible_stop
if not noval or final_epoch: # Calculate mAP
results, maps, _ = validate.run(data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
half=amp,
model=ema.ema,
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
plots=False,
callbacks=callbacks,
compute_loss=compute_loss)
# Update best mAP
fi = fitness(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
stop = stopper(epoch=epoch, fitness=fi) # early stop check
if fi > best_fitness:
best_fitness = fi
log_vals = list(mloss) + list(results) + lr
callbacks.run('on_fit_epoch_end', log_vals, epoch, best_fitness, fi)
# Save model
if (not nosave) or (final_epoch and not evolve): # if save
ckpt = {
'epoch': epoch,
'best_fitness': best_fitness,
'model': deepcopy(de_parallel(model)).half(),
'ema': deepcopy(ema.ema).half(),
'updates': ema.updates,
'optimizer': optimizer.state_dict(),
'opt': vars(opt),
'git': GIT_INFO, # {remote, branch, commit} if a git repo
'date': datetime.now().isoformat()}
# Save last, best and delete
torch.save(ckpt, last)
if best_fitness == fi:
torch.save(ckpt, best)
if opt.save_period > 0 and epoch % opt.save_period == 0:
torch.save(ckpt, w / f'epoch{epoch}.pt')
del ckpt
callbacks.run('on_model_save', last, epoch, final_epoch, best_fitness, fi)
# EarlyStopping
if RANK != -1: # if DDP training
broadcast_list = [stop if RANK == 0 else None]
dist.broadcast_object_list(broadcast_list, 0) # broadcast 'stop' to all ranks
if RANK != 0:
stop = broadcast_list[0]
if stop:
break # must break all DDP ranks
# end epoch ----------------------------------------------------------------------------------------------------
# end training -----------------------------------------------------------------------------------------------------
if RANK in {-1, 0}:
LOGGER.info(f'\n{epoch - start_epoch + 1} epochs completed in {(time.time() - t0) / 3600:.3f} hours.')
for f in last, best:
if f.exists():
strip_optimizer(f) # strip optimizers
if f is best:
LOGGER.info(f'\nValidating {f}...')
results, _, _ = validate.run(
data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
model=attempt_load(f, device).half(),
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
save_json=is_coco,
verbose=True,
plots=plots,
callbacks=callbacks,
compute_loss=compute_loss) # val best model with plots
if is_coco:
callbacks.run('on_fit_epoch_end', list(mloss) + list(results) + lr, epoch, best_fitness, fi)
callbacks.run('on_train_end', last, best, epoch, results)
torch.cuda.empty_cache()
return results
def parse_opt(known=False):
parser = argparse.ArgumentParser()
# parser.add_argument('--weights', type=str, default=ROOT / 'yolo.pt', help='initial weights path')
# parser.add_argument('--cfg', type=str, default='', help='model.yaml path')
parser.add_argument('--weights', type=str, default='', help='initial weights path')
parser.add_argument('--cfg', type=str, default='yolo.yaml', help='model.yaml path')
parser.add_argument('--data', type=str, default=ROOT / 'data/coco.yaml', help='dataset.yaml path')
parser.add_argument('--hyp', type=str, default=ROOT / 'data/hyps/hyp.scratch-high.yaml', help='hyperparameters path')
parser.add_argument('--epochs', type=int, default=100, help='total training epochs')
parser.add_argument('--batch-size', type=int, default=16, help='total batch size for all GPUs, -1 for autobatch')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='train, val image size (pixels)')
parser.add_argument('--rect', action='store_true', help='rectangular training')
parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
parser.add_argument('--noval', action='store_true', help='only validate final epoch')
parser.add_argument('--noautoanchor', action='store_true', help='disable AutoAnchor')
parser.add_argument('--noplots', action='store_true', help='save no plot files')
parser.add_argument('--evolve', type=int, nargs='?', const=300, help='evolve hyperparameters for x generations')
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
parser.add_argument('--cache', type=str, nargs='?', const='ram', help='image --cache ram/disk')
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
parser.add_argument('--single-cls', action='store_true', help='train multi-class data as single-class')
parser.add_argument('--optimizer', type=str, choices=['SGD', 'Adam', 'AdamW', 'LION'], default='SGD', help='optimizer')
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
parser.add_argument('--project', default=ROOT / 'runs/train', help='save to project/name')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--quad', action='store_true', help='quad dataloader')
parser.add_argument('--cos-lr', action='store_true', help='cosine LR scheduler')
parser.add_argument('--flat-cos-lr', action='store_true', help='flat cosine LR scheduler')
parser.add_argument('--fixed-lr', action='store_true', help='fixed LR scheduler')
parser.add_argument('--label-smoothing', type=float, default=0.0, help='Label smoothing epsilon')
parser.add_argument('--patience', type=int, default=100, help='EarlyStopping patience (epochs without improvement)')
parser.add_argument('--freeze', nargs='+', type=int, default=[0], help='Freeze layers: backbone=10, first3=0 1 2')
parser.add_argument('--save-period', type=int, default=-1, help='Save checkpoint every x epochs (disabled if < 1)')
parser.add_argument('--seed', type=int, default=0, help='Global training seed')
parser.add_argument('--local_rank', type=int, default=-1, help='Automatic DDP Multi-GPU argument, do not modify')
parser.add_argument('--min-items', type=int, default=0, help='Experimental')
parser.add_argument('--close-mosaic', type=int, default=0, help='Experimental')
# Logger arguments
parser.add_argument('--entity', default=None, help='Entity')
parser.add_argument('--upload_dataset', nargs='?', const=True, default=False, help='Upload data, "val" option')
parser.add_argument('--bbox_interval', type=int, default=-1, help='Set bounding-box image logging interval')
parser.add_argument('--artifact_alias', type=str, default='latest', help='Version of dataset artifact to use')
return parser.parse_known_args()[0] if known else parser.parse_args()
def main(opt, callbacks=Callbacks()):
# Checks
if RANK in {-1, 0}:
print_args(vars(opt))
#check_git_status()
#check_requirements()
# Resume (from specified or most recent last.pt)
if opt.resume and not check_comet_resume(opt) and not opt.evolve:
last = Path(check_file(opt.resume) if isinstance(opt.resume, str) else get_latest_run())
opt_yaml = last.parent.parent / 'opt.yaml' # train options yaml
opt_data = opt.data # original dataset
if opt_yaml.is_file():
with open(opt_yaml, errors='ignore') as f:
d = yaml.safe_load(f)
else:
d = torch.load(last, map_location='cpu')['opt']
opt = argparse.Namespace(**d) # replace
opt.cfg, opt.weights, opt.resume = '', str(last), True # reinstate
if is_url(opt_data):
opt.data = check_file(opt_data) # avoid HUB resume auth timeout
else:
opt.data, opt.cfg, opt.hyp, opt.weights, opt.project = \
check_file(opt.data), check_yaml(opt.cfg), check_yaml(opt.hyp), str(opt.weights), str(opt.project) # checks
assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
if opt.evolve:
if opt.project == str(ROOT / 'runs/train'): # if default project name, rename to runs/evolve
opt.project = str(ROOT / 'runs/evolve')
opt.exist_ok, opt.resume = opt.resume, False # pass resume to exist_ok and disable resume
if opt.name == 'cfg':
opt.name = Path(opt.cfg).stem # use model.yaml as name
opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok))
# DDP mode
device = select_device(opt.device, batch_size=opt.batch_size)
if LOCAL_RANK != -1:
msg = 'is not compatible with YOLO Multi-GPU DDP training'
assert not opt.image_weights, f'--image-weights {msg}'
assert not opt.evolve, f'--evolve {msg}'
assert opt.batch_size != -1, f'AutoBatch with --batch-size -1 {msg}, please pass a valid --batch-size'
assert opt.batch_size % WORLD_SIZE == 0, f'--batch-size {opt.batch_size} must be multiple of WORLD_SIZE'
assert torch.cuda.device_count() > LOCAL_RANK, 'insufficient CUDA devices for DDP command'
torch.cuda.set_device(LOCAL_RANK)
device = torch.device('cuda', LOCAL_RANK)
dist.init_process_group(backend="nccl" if dist.is_nccl_available() else "gloo")
# Train
if not opt.evolve:
train(opt.hyp, opt, device, callbacks)
# Evolve hyperparameters (optional)
else:
# Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
meta = {
'lr0': (1, 1e-5, 1e-1), # initial learning rate (SGD=1E-2, Adam=1E-3)
'lrf': (1, 0.01, 1.0), # final OneCycleLR learning rate (lr0 * lrf)
'momentum': (0.3, 0.6, 0.98), # SGD momentum/Adam beta1
'weight_decay': (1, 0.0, 0.001), # optimizer weight decay
'warmup_epochs': (1, 0.0, 5.0), # warmup epochs (fractions ok)
'warmup_momentum': (1, 0.0, 0.95), # warmup initial momentum
'warmup_bias_lr': (1, 0.0, 0.2), # warmup initial bias lr
'box': (1, 0.02, 0.2), # box loss gain
'cls': (1, 0.2, 4.0), # cls loss gain
'cls_pw': (1, 0.5, 2.0), # cls BCELoss positive_weight
'obj': (1, 0.2, 4.0), # obj loss gain (scale with pixels)
'obj_pw': (1, 0.5, 2.0), # obj BCELoss positive_weight
'iou_t': (0, 0.1, 0.7), # IoU training threshold
'anchor_t': (1, 2.0, 8.0), # anchor-multiple threshold
'anchors': (2, 2.0, 10.0), # anchors per output grid (0 to ignore)
'fl_gamma': (0, 0.0, 2.0), # focal loss gamma (efficientDet default gamma=1.5)
'hsv_h': (1, 0.0, 0.1), # image HSV-Hue augmentation (fraction)
'hsv_s': (1, 0.0, 0.9), # image HSV-Saturation augmentation (fraction)
'hsv_v': (1, 0.0, 0.9), # image HSV-Value augmentation (fraction)
'degrees': (1, 0.0, 45.0), # image rotation (+/- deg)
'translate': (1, 0.0, 0.9), # image translation (+/- fraction)
'scale': (1, 0.0, 0.9), # image scale (+/- gain)
'shear': (1, 0.0, 10.0), # image shear (+/- deg)
'perspective': (0, 0.0, 0.001), # image perspective (+/- fraction), range 0-0.001
'flipud': (1, 0.0, 1.0), # image flip up-down (probability)
'fliplr': (0, 0.0, 1.0), # image flip left-right (probability)
'mosaic': (1, 0.0, 1.0), # image mixup (probability)
'mixup': (1, 0.0, 1.0), # image mixup (probability)
'copy_paste': (1, 0.0, 1.0)} # segment copy-paste (probability)
with open(opt.hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
if 'anchors' not in hyp: # anchors commented in hyp.yaml
hyp['anchors'] = 3
if opt.noautoanchor:
del hyp['anchors'], meta['anchors']
opt.noval, opt.nosave, save_dir = True, True, Path(opt.save_dir) # only val/save final epoch
# ei = [isinstance(x, (int, float)) for x in hyp.values()] # evolvable indices
evolve_yaml, evolve_csv = save_dir / 'hyp_evolve.yaml', save_dir / 'evolve.csv'
if opt.bucket:
os.system(f'gsutil cp gs://{opt.bucket}/evolve.csv {evolve_csv}') # download evolve.csv if exists
for _ in range(opt.evolve): # generations to evolve
if evolve_csv.exists(): # if evolve.csv exists: select best hyps and mutate
# Select parent(s)
parent = 'single' # parent selection method: 'single' or 'weighted'
x = np.loadtxt(evolve_csv, ndmin=2, delimiter=',', skiprows=1)
n = min(5, len(x)) # number of previous results to consider
x = x[np.argsort(-fitness(x))][:n] # top n mutations
w = fitness(x) - fitness(x).min() + 1E-6 # weights (sum > 0)
if parent == 'single' or len(x) == 1:
# x = x[random.randint(0, n - 1)] # random selection
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
elif parent == 'weighted':
x = (x * w.reshape(n, 1)).sum(0) / w.sum() # weighted combination
# Mutate
mp, s = 0.8, 0.2 # mutation probability, sigma
npr = np.random
npr.seed(int(time.time()))
g = np.array([meta[k][0] for k in hyp.keys()]) # gains 0-1
ng = len(meta)
v = np.ones(ng)
while all(v == 1): # mutate until a change occurs (prevent duplicates)
v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
for i, k in enumerate(hyp.keys()): # plt.hist(v.ravel(), 300)
hyp[k] = float(x[i + 7] * v[i]) # mutate
# Constrain to limits
for k, v in meta.items():
hyp[k] = max(hyp[k], v[1]) # lower limit
hyp[k] = min(hyp[k], v[2]) # upper limit
hyp[k] = round(hyp[k], 5) # significant digits
# Train mutation
results = train(hyp.copy(), opt, device, callbacks)
callbacks = Callbacks()
# Write mutation results
keys = ('metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95', 'val/box_loss',
'val/obj_loss', 'val/cls_loss')
print_mutation(keys, results, hyp.copy(), save_dir, opt.bucket)
# Plot results
plot_evolve(evolve_csv)
LOGGER.info(f'Hyperparameter evolution finished {opt.evolve} generations\n'
f"Results saved to {colorstr('bold', save_dir)}\n"
f'Usage example: $ python train.py --hyp {evolve_yaml}')
def run(**kwargs):
# Usage: import train; train.run(data='coco128.yaml', imgsz=320, weights='yolo.pt')
opt = parse_opt(True)
for k, v in kwargs.items():
setattr(opt, k, v)
main(opt)
return opt
if __name__ == "__main__":
opt = parse_opt()
main(opt)
train_triple.py 0 → 100644
View file @ f969ca34
import argparse
import math
import os
import random
import sys
import time
from copy import deepcopy
from datetime import datetime
from pathlib import Path
import numpy as np
import torch
import torch.distributed as dist
import torch.nn as nn
import yaml
from torch.optim import lr_scheduler
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[0] # YOLO root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
import val_triple as validate # for end-of-epoch mAP
from models.experimental import attempt_load
from models.yolo import Model
from utils.autoanchor import check_anchors
from utils.autobatch import check_train_batch_size
from utils.callbacks import Callbacks
from utils.dataloaders import create_dataloader
from utils.downloads import attempt_download, is_url
from utils.general import (LOGGER, TQDM_BAR_FORMAT, check_amp, check_dataset, check_file, check_git_info,
check_git_status, check_img_size, check_requirements, check_suffix, check_yaml, colorstr,
get_latest_run, increment_path, init_seeds, intersect_dicts, labels_to_class_weights,
labels_to_image_weights, methods, one_cycle, print_args, print_mutation, strip_optimizer,
yaml_save)
from utils.loggers import Loggers
from utils.loggers.comet.comet_utils import check_comet_resume
from utils.loss_tal_triple import ComputeLoss
from utils.metrics import fitness
from utils.plots import plot_evolve
from utils.torch_utils import (EarlyStopping, ModelEMA, de_parallel, select_device, smart_DDP, smart_optimizer,
smart_resume, torch_distributed_zero_first)
LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1)) # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
GIT_INFO = None#check_git_info()
def train(hyp, opt, device, callbacks): # hyp is path/to/hyp.yaml or hyp dictionary
save_dir, epochs, batch_size, weights, single_cls, evolve, data, cfg, resume, noval, nosave, workers, freeze = \
Path(opt.save_dir), opt.epochs, opt.batch_size, opt.weights, opt.single_cls, opt.evolve, opt.data, opt.cfg, \
opt.resume, opt.noval, opt.nosave, opt.workers, opt.freeze
callbacks.run('on_pretrain_routine_start')
# Directories
w = save_dir / 'weights' # weights dir
(w.parent if evolve else w).mkdir(parents=True, exist_ok=True) # make dir
last, best = w / 'last.pt', w / 'best.pt'
# Hyperparameters
if isinstance(hyp, str):
with open(hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
LOGGER.info(colorstr('hyperparameters: ') + ', '.join(f'{k}={v}' for k, v in hyp.items()))
hyp['anchor_t'] = 5.0
opt.hyp = hyp.copy() # for saving hyps to checkpoints
# Save run settings
if not evolve:
yaml_save(save_dir / 'hyp.yaml', hyp)
yaml_save(save_dir / 'opt.yaml', vars(opt))
# Loggers
data_dict = None
if RANK in {-1, 0}:
loggers = Loggers(save_dir, weights, opt, hyp, LOGGER) # loggers instance
# Register actions
for k in methods(loggers):
callbacks.register_action(k, callback=getattr(loggers, k))
# Process custom dataset artifact link
data_dict = loggers.remote_dataset
if resume: # If resuming runs from remote artifact
weights, epochs, hyp, batch_size = opt.weights, opt.epochs, opt.hyp, opt.batch_size
# Config
plots = not evolve and not opt.noplots # create plots
cuda = device.type != 'cpu'
init_seeds(opt.seed + 1 + RANK, deterministic=True)
with torch_distributed_zero_first(LOCAL_RANK):
data_dict = data_dict or check_dataset(data) # check if None
train_path, val_path = data_dict['train'], data_dict['val']
nc = 1 if single_cls else int(data_dict['nc']) # number of classes
names = {0: 'item'} if single_cls and len(data_dict['names']) != 1 else data_dict['names'] # class names
#is_coco = isinstance(val_path, str) and val_path.endswith('coco/val2017.txt') # COCO dataset
is_coco = isinstance(val_path, str) and val_path.endswith('val2017.txt') # COCO dataset
# Model
check_suffix(weights, '.pt') # check weights
pretrained = weights.endswith('.pt')
if pretrained:
with torch_distributed_zero_first(LOCAL_RANK):
weights = attempt_download(weights) # download if not found locally
ckpt = torch.load(weights, map_location='cpu') # load checkpoint to CPU to avoid CUDA memory leak
model = Model(cfg or ckpt['model'].yaml, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # create
exclude = ['anchor'] if (cfg or hyp.get('anchors')) and not resume else [] # exclude keys
csd = ckpt['model'].float().state_dict() # checkpoint state_dict as FP32
csd = intersect_dicts(csd, model.state_dict(), exclude=exclude) # intersect
model.load_state_dict(csd, strict=False) # load
LOGGER.info(f'Transferred {len(csd)}/{len(model.state_dict())} items from {weights}') # report
else:
model = Model(cfg, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # create
amp = check_amp(model) # check AMP
# Freeze
freeze = [f'model.{x}.' for x in (freeze if len(freeze) > 1 else range(freeze[0]))] # layers to freeze
for k, v in model.named_parameters():
# v.requires_grad = True # train all layers TODO: uncomment this line as in master
# v.register_hook(lambda x: torch.nan_to_num(x)) # NaN to 0 (commented for erratic training results)
if any(x in k for x in freeze):
LOGGER.info(f'freezing {k}')
v.requires_grad = False
# Image size
gs = max(int(model.stride.max()), 32) # grid size (max stride)
imgsz = check_img_size(opt.imgsz, gs, floor=gs * 2) # verify imgsz is gs-multiple
# Batch size
if RANK == -1 and batch_size == -1: # single-GPU only, estimate best batch size
batch_size = check_train_batch_size(model, imgsz, amp)
loggers.on_params_update({"batch_size": batch_size})
# Optimizer
nbs = 64 # nominal batch size
accumulate = max(round(nbs / batch_size), 1) # accumulate loss before optimizing
hyp['weight_decay'] *= batch_size * accumulate / nbs # scale weight_decay
optimizer = smart_optimizer(model, opt.optimizer, hyp['lr0'], hyp['momentum'], hyp['weight_decay'])
# Scheduler
if opt.cos_lr:
lf = one_cycle(1, hyp['lrf'], epochs) # cosine 1->hyp['lrf']
else:
lf = lambda x: (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf'] # linear
# def lf(x): # saw
# return (1 - (x % 30) / 30) * (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf']
#
# def lf(x): # triangle start at min
# return 2 * abs(x / 30 - math.floor(x / 30 + 1 / 2)) * (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf']
#
# def lf(x): # triangle start at max
# return 2 * abs(x / 32 + .5 - math.floor(x / 32 + 1)) * (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf']
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf)
# from utils.plots import plot_lr_scheduler; plot_lr_scheduler(optimizer, scheduler, epochs)
# EMA
ema = ModelEMA(model) if RANK in {-1, 0} else None
# Resume
best_fitness, start_epoch = 0.0, 0
if pretrained:
if resume:
best_fitness, start_epoch, epochs = smart_resume(ckpt, optimizer, ema, weights, epochs, resume)
del ckpt, csd
# DP mode
if cuda and RANK == -1 and torch.cuda.device_count() > 1:
LOGGER.warning('WARNING ⚠️ DP not recommended, use torch.distributed.run for best DDP Multi-GPU results.')
model = torch.nn.DataParallel(model)
# SyncBatchNorm
if opt.sync_bn and cuda and RANK != -1:
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
LOGGER.info('Using SyncBatchNorm()')
# Trainloader
train_loader, dataset = create_dataloader(train_path,
imgsz,
batch_size // WORLD_SIZE,
gs,
single_cls,
hyp=hyp,
augment=True,
cache=None if opt.cache == 'val' else opt.cache,
rect=opt.rect,
rank=LOCAL_RANK,
workers=workers,
image_weights=opt.image_weights,
close_mosaic=opt.close_mosaic != 0,
quad=opt.quad,
prefix=colorstr('train: '),
shuffle=True,
min_items=opt.min_items)
labels = np.concatenate(dataset.labels, 0)
mlc = int(labels[:, 0].max()) # max label class
assert mlc < nc, f'Label class {mlc} exceeds nc={nc} in {data}. Possible class labels are 0-{nc - 1}'
# Process 0
if RANK in {-1, 0}:
val_loader = create_dataloader(val_path,
imgsz,
batch_size // WORLD_SIZE * 2,
gs,
single_cls,
hyp=hyp,
cache=None if noval else opt.cache,
rect=True,
rank=-1,
workers=workers * 2,
pad=0.5,
prefix=colorstr('val: '))[0]
if not resume:
# if not opt.noautoanchor:
# check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz) # run AutoAnchor
model.half().float() # pre-reduce anchor precision
callbacks.run('on_pretrain_routine_end', labels, names)
# DDP mode
if cuda and RANK != -1:
model = smart_DDP(model)
# Model attributes
nl = de_parallel(model).model[-1].nl # number of detection layers (to scale hyps)
#hyp['box'] *= 3 / nl # scale to layers
#hyp['cls'] *= nc / 80 * 3 / nl # scale to classes and layers
#hyp['obj'] *= (imgsz / 640) ** 2 * 3 / nl # scale to image size and layers
hyp['label_smoothing'] = opt.label_smoothing
model.nc = nc # attach number of classes to model
model.hyp = hyp # attach hyperparameters to model
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) * nc # attach class weights
model.names = names
# Start training
t0 = time.time()
nb = len(train_loader) # number of batches
nw = max(round(hyp['warmup_epochs'] * nb), 100) # number of warmup iterations, max(3 epochs, 100 iterations)
# nw = min(nw, (epochs - start_epoch) / 2 * nb) # limit warmup to < 1/2 of training
last_opt_step = -1
maps = np.zeros(nc) # mAP per class
results = (0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
scheduler.last_epoch = start_epoch - 1 # do not move
scaler = torch.cuda.amp.GradScaler(enabled=amp)
stopper, stop = EarlyStopping(patience=opt.patience), False
compute_loss = ComputeLoss(model) # init loss class
callbacks.run('on_train_start')
LOGGER.info(f'Image sizes {imgsz} train, {imgsz} val\n'
f'Using {train_loader.num_workers * WORLD_SIZE} dataloader workers\n'
f"Logging results to {colorstr('bold', save_dir)}\n"
f'Starting training for {epochs} epochs...')
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
callbacks.run('on_train_epoch_start')
model.train()
# Update image weights (optional, single-GPU only)
if opt.image_weights:
cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 / nc # class weights
iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw) # image weights
dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n) # rand weighted idx
if epoch == (epochs - opt.close_mosaic):
LOGGER.info("Closing dataloader mosaic")
dataset.mosaic = False
# Update mosaic border (optional)
# b = int(random.uniform(0.25 * imgsz, 0.75 * imgsz + gs) // gs * gs)
# dataset.mosaic_border = [b - imgsz, -b] # height, width borders
mloss = torch.zeros(3, device=device) # mean losses
if RANK != -1:
train_loader.sampler.set_epoch(epoch)
pbar = enumerate(train_loader)
LOGGER.info(('\n' + '%11s' * 7) % ('Epoch', 'GPU_mem', 'box_loss', 'cls_loss', 'dfl_loss', 'Instances', 'Size'))
if RANK in {-1, 0}:
pbar = tqdm(pbar, total=nb, bar_format=TQDM_BAR_FORMAT) # progress bar
optimizer.zero_grad()
for i, (imgs, targets, paths, _) in pbar: # batch -------------------------------------------------------------
callbacks.run('on_train_batch_start')
ni = i + nb * epoch # number integrated batches (since train start)
imgs = imgs.to(device, non_blocking=True).float() / 255 # uint8 to float32, 0-255 to 0.0-1.0
# Warmup
if ni <= nw:
xi = [0, nw] # x interp
# compute_loss.gr = np.interp(ni, xi, [0.0, 1.0]) # iou loss ratio (obj_loss = 1.0 or iou)
accumulate = max(1, np.interp(ni, xi, [1, nbs / batch_size]).round())
for j, x in enumerate(optimizer.param_groups):
# bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
x['lr'] = np.interp(ni, xi, [hyp['warmup_bias_lr'] if j == 0 else 0.0, x['initial_lr'] * lf(epoch)])
if 'momentum' in x:
x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])
# Multi-scale
if opt.multi_scale:
sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs # size
sf = sz / max(imgs.shape[2:]) # scale factor
if sf != 1:
ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]] # new shape (stretched to gs-multiple)
imgs = nn.functional.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)
# Forward
with torch.cuda.amp.autocast(amp):
pred = model(imgs) # forward
loss, loss_items = compute_loss(pred, targets.to(device)) # loss scaled by batch_size
if RANK != -1:
loss *= WORLD_SIZE # gradient averaged between devices in DDP mode
if opt.quad:
loss *= 4.
# Backward
scaler.scale(loss).backward()
# Optimize - https://pytorch.org/docs/master/notes/amp_examples.html
if ni - last_opt_step >= accumulate:
scaler.unscale_(optimizer) # unscale gradients
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=10.0) # clip gradients
scaler.step(optimizer) # optimizer.step
scaler.update()
optimizer.zero_grad()
if ema:
ema.update(model)
last_opt_step = ni
# Log
if RANK in {-1, 0}:
mloss = (mloss * i + loss_items) / (i + 1) # update mean losses
mem = f'{torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0:.3g}G' # (GB)
pbar.set_description(('%11s' * 2 + '%11.4g' * 5) %
(f'{epoch}/{epochs - 1}', mem, *mloss, targets.shape[0], imgs.shape[-1]))
callbacks.run('on_train_batch_end', model, ni, imgs, targets, paths, list(mloss))
if callbacks.stop_training:
return
# end batch ------------------------------------------------------------------------------------------------
# Scheduler
lr = [x['lr'] for x in optimizer.param_groups] # for loggers
scheduler.step()
if RANK in {-1, 0}:
# mAP
callbacks.run('on_train_epoch_end', epoch=epoch)
ema.update_attr(model, include=['yaml', 'nc', 'hyp', 'names', 'stride', 'class_weights'])
final_epoch = (epoch + 1 == epochs) or stopper.possible_stop
if not noval or final_epoch: # Calculate mAP
results, maps, _ = validate.run(data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
half=amp,
model=ema.ema,
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
plots=False,
callbacks=callbacks,
compute_loss=compute_loss)
# Update best mAP
fi = fitness(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
stop = stopper(epoch=epoch, fitness=fi) # early stop check
if fi > best_fitness:
best_fitness = fi
log_vals = list(mloss) + list(results) + lr
callbacks.run('on_fit_epoch_end', log_vals, epoch, best_fitness, fi)
# Save model
if (not nosave) or (final_epoch and not evolve): # if save
ckpt = {
'epoch': epoch,
'best_fitness': best_fitness,
'model': deepcopy(de_parallel(model)).half(),
'ema': deepcopy(ema.ema).half(),
'updates': ema.updates,
'optimizer': optimizer.state_dict(),
'opt': vars(opt),
'git': GIT_INFO, # {remote, branch, commit} if a git repo
'date': datetime.now().isoformat()}
# Save last, best and delete
torch.save(ckpt, last)
if best_fitness == fi:
torch.save(ckpt, best)
if opt.save_period > 0 and epoch % opt.save_period == 0:
torch.save(ckpt, w / f'epoch{epoch}.pt')
del ckpt
callbacks.run('on_model_save', last, epoch, final_epoch, best_fitness, fi)
# EarlyStopping
if RANK != -1: # if DDP training
broadcast_list = [stop if RANK == 0 else None]
dist.broadcast_object_list(broadcast_list, 0) # broadcast 'stop' to all ranks
if RANK != 0:
stop = broadcast_list[0]
if stop:
break # must break all DDP ranks
# end epoch ----------------------------------------------------------------------------------------------------
# end training -----------------------------------------------------------------------------------------------------
if RANK in {-1, 0}:
LOGGER.info(f'\n{epoch - start_epoch + 1} epochs completed in {(time.time() - t0) / 3600:.3f} hours.')
for f in last, best:
if f.exists():
strip_optimizer(f) # strip optimizers
if f is best:
LOGGER.info(f'\nValidating {f}...')
results, _, _ = validate.run(
data_dict,
batch_size=batch_size // WORLD_SIZE * 2,
imgsz=imgsz,
model=attempt_load(f, device).half(),
single_cls=single_cls,
dataloader=val_loader,
save_dir=save_dir,
save_json=is_coco,
verbose=True,
plots=plots,
callbacks=callbacks,
compute_loss=compute_loss) # val best model with plots
if is_coco:
callbacks.run('on_fit_epoch_end', list(mloss) + list(results) + lr, epoch, best_fitness, fi)
callbacks.run('on_train_end', last, best, epoch, results)
torch.cuda.empty_cache()
return results
def parse_opt(known=False):
parser = argparse.ArgumentParser()
# parser.add_argument('--weights', type=str, default=ROOT / 'yolo.pt', help='initial weights path')
# parser.add_argument('--cfg', type=str, default='', help='model.yaml path')
parser.add_argument('--weights', type=str, default='', help='initial weights path')
parser.add_argument('--cfg', type=str, default='yolo.yaml', help='model.yaml path')
parser.add_argument('--data', type=str, default=ROOT / 'data/coco.yaml', help='dataset.yaml path')
parser.add_argument('--hyp', type=str, default=ROOT / 'data/hyps/hyp.scratch-high.yaml', help='hyperparameters path')
parser.add_argument('--epochs', type=int, default=100, help='total training epochs')
parser.add_argument('--batch-size', type=int, default=16, help='total batch size for all GPUs, -1 for autobatch')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='train, val image size (pixels)')
parser.add_argument('--rect', action='store_true', help='rectangular training')
parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
parser.add_argument('--noval', action='store_true', help='only validate final epoch')
parser.add_argument('--noautoanchor', action='store_true', help='disable AutoAnchor')
parser.add_argument('--noplots', action='store_true', help='save no plot files')
parser.add_argument('--evolve', type=int, nargs='?', const=300, help='evolve hyperparameters for x generations')
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
parser.add_argument('--cache', type=str, nargs='?', const='ram', help='image --cache ram/disk')
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
parser.add_argument('--single-cls', action='store_true', help='train multi-class data as single-class')
parser.add_argument('--optimizer', type=str, choices=['SGD', 'Adam', 'AdamW', 'LION'], default='SGD', help='optimizer')
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)')
parser.add_argument('--project', default=ROOT / 'runs/train', help='save to project/name')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--quad', action='store_true', help='quad dataloader')
parser.add_argument('--cos-lr', action='store_true', help='cosine LR scheduler')
parser.add_argument('--label-smoothing', type=float, default=0.0, help='Label smoothing epsilon')
parser.add_argument('--patience', type=int, default=100, help='EarlyStopping patience (epochs without improvement)')
parser.add_argument('--freeze', nargs='+', type=int, default=[0], help='Freeze layers: backbone=10, first3=0 1 2')
parser.add_argument('--save-period', type=int, default=-1, help='Save checkpoint every x epochs (disabled if < 1)')
parser.add_argument('--seed', type=int, default=0, help='Global training seed')
parser.add_argument('--local_rank', type=int, default=-1, help='Automatic DDP Multi-GPU argument, do not modify')
parser.add_argument('--min-items', type=int, default=0, help='Experimental')
parser.add_argument('--close-mosaic', type=int, default=0, help='Experimental')
# Logger arguments
parser.add_argument('--entity', default=None, help='Entity')
parser.add_argument('--upload_dataset', nargs='?', const=True, default=False, help='Upload data, "val" option')
parser.add_argument('--bbox_interval', type=int, default=-1, help='Set bounding-box image logging interval')
parser.add_argument('--artifact_alias', type=str, default='latest', help='Version of dataset artifact to use')
return parser.parse_known_args()[0] if known else parser.parse_args()
def main(opt, callbacks=Callbacks()):
# Checks
if RANK in {-1, 0}:
print_args(vars(opt))
#check_git_status()
#check_requirements()
# Resume (from specified or most recent last.pt)
if opt.resume and not check_comet_resume(opt) and not opt.evolve:
last = Path(check_file(opt.resume) if isinstance(opt.resume, str) else get_latest_run())
opt_yaml = last.parent.parent / 'opt.yaml' # train options yaml
opt_data = opt.data # original dataset
if opt_yaml.is_file():
with open(opt_yaml, errors='ignore') as f:
d = yaml.safe_load(f)
else:
d = torch.load(last, map_location='cpu')['opt']
opt = argparse.Namespace(**d) # replace
opt.cfg, opt.weights, opt.resume = '', str(last), True # reinstate
if is_url(opt_data):
opt.data = check_file(opt_data) # avoid HUB resume auth timeout
else:
opt.data, opt.cfg, opt.hyp, opt.weights, opt.project = \
check_file(opt.data), check_yaml(opt.cfg), check_yaml(opt.hyp), str(opt.weights), str(opt.project) # checks
assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
if opt.evolve:
if opt.project == str(ROOT / 'runs/train'): # if default project name, rename to runs/evolve
opt.project = str(ROOT / 'runs/evolve')
opt.exist_ok, opt.resume = opt.resume, False # pass resume to exist_ok and disable resume
if opt.name == 'cfg':
opt.name = Path(opt.cfg).stem # use model.yaml as name
opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok))
# DDP mode
device = select_device(opt.device, batch_size=opt.batch_size)
if LOCAL_RANK != -1:
msg = 'is not compatible with YOLO Multi-GPU DDP training'
assert not opt.image_weights, f'--image-weights {msg}'
assert not opt.evolve, f'--evolve {msg}'
assert opt.batch_size != -1, f'AutoBatch with --batch-size -1 {msg}, please pass a valid --batch-size'
assert opt.batch_size % WORLD_SIZE == 0, f'--batch-size {opt.batch_size} must be multiple of WORLD_SIZE'
assert torch.cuda.device_count() > LOCAL_RANK, 'insufficient CUDA devices for DDP command'
torch.cuda.set_device(LOCAL_RANK)
device = torch.device('cuda', LOCAL_RANK)
dist.init_process_group(backend="nccl" if dist.is_nccl_available() else "gloo")
# Train
if not opt.evolve:
train(opt.hyp, opt, device, callbacks)
# Evolve hyperparameters (optional)
else:
# Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
meta = {
'lr0': (1, 1e-5, 1e-1), # initial learning rate (SGD=1E-2, Adam=1E-3)
'lrf': (1, 0.01, 1.0), # final OneCycleLR learning rate (lr0 * lrf)
'momentum': (0.3, 0.6, 0.98), # SGD momentum/Adam beta1
'weight_decay': (1, 0.0, 0.001), # optimizer weight decay
'warmup_epochs': (1, 0.0, 5.0), # warmup epochs (fractions ok)
'warmup_momentum': (1, 0.0, 0.95), # warmup initial momentum
'warmup_bias_lr': (1, 0.0, 0.2), # warmup initial bias lr
'box': (1, 0.02, 0.2), # box loss gain
'cls': (1, 0.2, 4.0), # cls loss gain
'cls_pw': (1, 0.5, 2.0), # cls BCELoss positive_weight
'obj': (1, 0.2, 4.0), # obj loss gain (scale with pixels)
'obj_pw': (1, 0.5, 2.0), # obj BCELoss positive_weight
'iou_t': (0, 0.1, 0.7), # IoU training threshold
'anchor_t': (1, 2.0, 8.0), # anchor-multiple threshold
'anchors': (2, 2.0, 10.0), # anchors per output grid (0 to ignore)
'fl_gamma': (0, 0.0, 2.0), # focal loss gamma (efficientDet default gamma=1.5)
'hsv_h': (1, 0.0, 0.1), # image HSV-Hue augmentation (fraction)
'hsv_s': (1, 0.0, 0.9), # image HSV-Saturation augmentation (fraction)
'hsv_v': (1, 0.0, 0.9), # image HSV-Value augmentation (fraction)
'degrees': (1, 0.0, 45.0), # image rotation (+/- deg)
'translate': (1, 0.0, 0.9), # image translation (+/- fraction)
'scale': (1, 0.0, 0.9), # image scale (+/- gain)
'shear': (1, 0.0, 10.0), # image shear (+/- deg)
'perspective': (0, 0.0, 0.001), # image perspective (+/- fraction), range 0-0.001
'flipud': (1, 0.0, 1.0), # image flip up-down (probability)
'fliplr': (0, 0.0, 1.0), # image flip left-right (probability)
'mosaic': (1, 0.0, 1.0), # image mixup (probability)
'mixup': (1, 0.0, 1.0), # image mixup (probability)
'copy_paste': (1, 0.0, 1.0)} # segment copy-paste (probability)
with open(opt.hyp, errors='ignore') as f:
hyp = yaml.safe_load(f) # load hyps dict
if 'anchors' not in hyp: # anchors commented in hyp.yaml
hyp['anchors'] = 3
if opt.noautoanchor:
del hyp['anchors'], meta['anchors']
opt.noval, opt.nosave, save_dir = True, True, Path(opt.save_dir) # only val/save final epoch
# ei = [isinstance(x, (int, float)) for x in hyp.values()] # evolvable indices
evolve_yaml, evolve_csv = save_dir / 'hyp_evolve.yaml', save_dir / 'evolve.csv'
if opt.bucket:
os.system(f'gsutil cp gs://{opt.bucket}/evolve.csv {evolve_csv}') # download evolve.csv if exists
for _ in range(opt.evolve): # generations to evolve
if evolve_csv.exists(): # if evolve.csv exists: select best hyps and mutate
# Select parent(s)
parent = 'single' # parent selection method: 'single' or 'weighted'
x = np.loadtxt(evolve_csv, ndmin=2, delimiter=',', skiprows=1)
n = min(5, len(x)) # number of previous results to consider
x = x[np.argsort(-fitness(x))][:n] # top n mutations
w = fitness(x) - fitness(x).min() + 1E-6 # weights (sum > 0)
if parent == 'single' or len(x) == 1:
# x = x[random.randint(0, n - 1)] # random selection
x = x[random.choices(range(n), weights=w)[0]] # weighted selection
elif parent == 'weighted':
x = (x * w.reshape(n, 1)).sum(0) / w.sum() # weighted combination
# Mutate
mp, s = 0.8, 0.2 # mutation probability, sigma
npr = np.random
npr.seed(int(time.time()))
g = np.array([meta[k][0] for k in hyp.keys()]) # gains 0-1
ng = len(meta)
v = np.ones(ng)
while all(v == 1): # mutate until a change occurs (prevent duplicates)
v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
for i, k in enumerate(hyp.keys()): # plt.hist(v.ravel(), 300)
hyp[k] = float(x[i + 7] * v[i]) # mutate
# Constrain to limits
for k, v in meta.items():
hyp[k] = max(hyp[k], v[1]) # lower limit
hyp[k] = min(hyp[k], v[2]) # upper limit
hyp[k] = round(hyp[k], 5) # significant digits
# Train mutation
results = train(hyp.copy(), opt, device, callbacks)
callbacks = Callbacks()
# Write mutation results
keys = ('metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95', 'val/box_loss',
'val/obj_loss', 'val/cls_loss')
print_mutation(keys, results, hyp.copy(), save_dir, opt.bucket)
# Plot results
plot_evolve(evolve_csv)
LOGGER.info(f'Hyperparameter evolution finished {opt.evolve} generations\n'
f"Results saved to {colorstr('bold', save_dir)}\n"
f'Usage example: $ python train.py --hyp {evolve_yaml}')
def run(**kwargs):
# Usage: import train; train.run(data='coco128.yaml', imgsz=320, weights='yolo.pt')
opt = parse_opt(True)
for k, v in kwargs.items():
setattr(opt, k, v)
main(opt)
return opt
if __name__ == "__main__":
opt = parse_opt()
main(opt)
utils/__init__.py 0 → 100644
View file @ f969ca34
import contextlib
import platform
import threading
def emojis(str=''):
# Return platform-dependent emoji-safe version of string
return str.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else str
class TryExcept(contextlib.ContextDecorator):
# YOLOv5 TryExcept class. Usage: @TryExcept() decorator or 'with TryExcept():' context manager
def __init__(self, msg=''):
self.msg = msg
def __enter__(self):
pass
def __exit__(self, exc_type, value, traceback):
if value:
print(emojis(f"{self.msg}{': ' if self.msg else ''}{value}"))
return True
def threaded(func):
# Multi-threads a target function and returns thread. Usage: @threaded decorator
def wrapper(*args, **kwargs):
thread = threading.Thread(target=func, args=args, kwargs=kwargs, daemon=True)
thread.start()
return thread
return wrapper
def join_threads(verbose=False):
# Join all daemon threads, i.e. atexit.register(lambda: join_threads())
main_thread = threading.current_thread()
for t in threading.enumerate():
if t is not main_thread:
if verbose:
print(f'Joining thread {t.name}')
t.join()
def notebook_init(verbose=True):
# Check system software and hardware
print('Checking setup...')
import os
import shutil
from utils.general import check_font, check_requirements, is_colab
from utils.torch_utils import select_device # imports
check_font()
import psutil
from IPython import display # to display images and clear console output
if is_colab():
shutil.rmtree('/content/sample_data', ignore_errors=True) # remove colab /sample_data directory
# System info
if verbose:
gb = 1 << 30 # bytes to GiB (1024 ** 3)
ram = psutil.virtual_memory().total
total, used, free = shutil.disk_usage("/")
display.clear_output()
s = f'({os.cpu_count()} CPUs, {ram / gb:.1f} GB RAM, {(total - free) / gb:.1f}/{total / gb:.1f} GB disk)'
else:
s = ''
select_device(newline=False)
print(emojis(f'Setup complete ✅ {s}'))
return display
utils/activations.py 0 → 100644
View file @ f969ca34
import torch
import torch.nn as nn
import torch.nn.functional as F
class SiLU(nn.Module):
# SiLU activation https://arxiv.org/pdf/1606.08415.pdf
@staticmethod
def forward(x):
return x * torch.sigmoid(x)
class Hardswish(nn.Module):
# Hard-SiLU activation
@staticmethod
def forward(x):
# return x * F.hardsigmoid(x) # for TorchScript and CoreML
return x * F.hardtanh(x + 3, 0.0, 6.0) / 6.0 # for TorchScript, CoreML and ONNX
class Mish(nn.Module):
# Mish activation https://github.com/digantamisra98/Mish
@staticmethod
def forward(x):
return x * F.softplus(x).tanh()
class MemoryEfficientMish(nn.Module):
# Mish activation memory-efficient
class F(torch.autograd.Function):
@staticmethod
def forward(ctx, x):
ctx.save_for_backward(x)
return x.mul(torch.tanh(F.softplus(x))) # x * tanh(ln(1 + exp(x)))
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_tensors[0]
sx = torch.sigmoid(x)
fx = F.softplus(x).tanh()
return grad_output * (fx + x * sx * (1 - fx * fx))
def forward(self, x):
return self.F.apply(x)
class FReLU(nn.Module):
# FReLU activation https://arxiv.org/abs/2007.11824
def __init__(self, c1, k=3): # ch_in, kernel
super().__init__()
self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1, bias=False)
self.bn = nn.BatchNorm2d(c1)
def forward(self, x):
return torch.max(x, self.bn(self.conv(x)))
class AconC(nn.Module):
r""" ACON activation (activate or not)
AconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is a learnable parameter
according to "Activate or Not: Learning Customized Activation" <https://arxiv.org/pdf/2009.04759.pdf>.
"""
def __init__(self, c1):
super().__init__()
self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1))
self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1))
self.beta = nn.Parameter(torch.ones(1, c1, 1, 1))
def forward(self, x):
dpx = (self.p1 - self.p2) * x
return dpx * torch.sigmoid(self.beta * dpx) + self.p2 * x
class MetaAconC(nn.Module):
r""" ACON activation (activate or not)
MetaAconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is generated by a small network
according to "Activate or Not: Learning Customized Activation" <https://arxiv.org/pdf/2009.04759.pdf>.
"""
def __init__(self, c1, k=1, s=1, r=16): # ch_in, kernel, stride, r
super().__init__()
c2 = max(r, c1 // r)
self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1))
self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1))
self.fc1 = nn.Conv2d(c1, c2, k, s, bias=True)
self.fc2 = nn.Conv2d(c2, c1, k, s, bias=True)
# self.bn1 = nn.BatchNorm2d(c2)
# self.bn2 = nn.BatchNorm2d(c1)
def forward(self, x):
y = x.mean(dim=2, keepdims=True).mean(dim=3, keepdims=True)
# batch-size 1 bug/instabilities https://github.com/ultralytics/yolov5/issues/2891
# beta = torch.sigmoid(self.bn2(self.fc2(self.bn1(self.fc1(y))))) # bug/unstable
beta = torch.sigmoid(self.fc2(self.fc1(y))) # bug patch BN layers removed
dpx = (self.p1 - self.p2) * x
return dpx * torch.sigmoid(beta * dpx) + self.p2 * x
utils/augmentations.py 0 → 100644
View file @ f969ca34
import math
import random
import cv2
import numpy as np
import torch
import torchvision.transforms as T
import torchvision.transforms.functional as TF
from utils.general import LOGGER, check_version, colorstr, resample_segments, segment2box, xywhn2xyxy
from utils.metrics import bbox_ioa
IMAGENET_MEAN = 0.485, 0.456, 0.406 # RGB mean
IMAGENET_STD = 0.229, 0.224, 0.225 # RGB standard deviation
class Albumentations:
# YOLOv5 Albumentations class (optional, only used if package is installed)
def __init__(self, size=640):
self.transform = None
prefix = colorstr('albumentations: ')
try:
import albumentations as A
check_version(A.__version__, '1.0.3', hard=True) # version requirement
T = [
A.RandomResizedCrop(height=size, width=size, scale=(0.8, 1.0), ratio=(0.9, 1.11), p=0.0),
A.Blur(p=0.01),
A.MedianBlur(p=0.01),
A.ToGray(p=0.01),
A.CLAHE(p=0.01),
A.RandomBrightnessContrast(p=0.0),
A.RandomGamma(p=0.0),
A.ImageCompression(quality_lower=75, p=0.0)] # transforms
self.transform = A.Compose(T, bbox_params=A.BboxParams(format='yolo', label_fields=['class_labels']))
LOGGER.info(prefix + ', '.join(f'{x}'.replace('always_apply=False, ', '') for x in T if x.p))
except ImportError: # package not installed, skip
pass
except Exception as e:
LOGGER.info(f'{prefix}{e}')
def __call__(self, im, labels, p=1.0):
if self.transform and random.random() < p:
new = self.transform(image=im, bboxes=labels[:, 1:], class_labels=labels[:, 0]) # transformed
im, labels = new['image'], np.array([[c, *b] for c, b in zip(new['class_labels'], new['bboxes'])])
return im, labels
def normalize(x, mean=IMAGENET_MEAN, std=IMAGENET_STD, inplace=False):
# Denormalize RGB images x per ImageNet stats in BCHW format, i.e. = (x - mean) / std
return TF.normalize(x, mean, std, inplace=inplace)
def denormalize(x, mean=IMAGENET_MEAN, std=IMAGENET_STD):
# Denormalize RGB images x per ImageNet stats in BCHW format, i.e. = x * std + mean
for i in range(3):
x[:, i] = x[:, i] * std[i] + mean[i]
return x
def augment_hsv(im, hgain=0.5, sgain=0.5, vgain=0.5):
# HSV color-space augmentation
if hgain or sgain or vgain:
r = np.random.uniform(-1, 1, 3) * [hgain, sgain, vgain] + 1 # random gains
hue, sat, val = cv2.split(cv2.cvtColor(im, cv2.COLOR_BGR2HSV))
dtype = im.dtype # uint8
x = np.arange(0, 256, dtype=r.dtype)
lut_hue = ((x * r[0]) % 180).astype(dtype)
lut_sat = np.clip(x * r[1], 0, 255).astype(dtype)
lut_val = np.clip(x * r[2], 0, 255).astype(dtype)
im_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val)))
cv2.cvtColor(im_hsv, cv2.COLOR_HSV2BGR, dst=im) # no return needed
def hist_equalize(im, clahe=True, bgr=False):
# Equalize histogram on BGR image 'im' with im.shape(n,m,3) and range 0-255
yuv = cv2.cvtColor(im, cv2.COLOR_BGR2YUV if bgr else cv2.COLOR_RGB2YUV)
if clahe:
c = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
yuv[:, :, 0] = c.apply(yuv[:, :, 0])
else:
yuv[:, :, 0] = cv2.equalizeHist(yuv[:, :, 0]) # equalize Y channel histogram
return cv2.cvtColor(yuv, cv2.COLOR_YUV2BGR if bgr else cv2.COLOR_YUV2RGB) # convert YUV image to RGB
def replicate(im, labels):
# Replicate labels
h, w = im.shape[:2]
boxes = labels[:, 1:].astype(int)
x1, y1, x2, y2 = boxes.T
s = ((x2 - x1) + (y2 - y1)) / 2 # side length (pixels)
for i in s.argsort()[:round(s.size * 0.5)]: # smallest indices
x1b, y1b, x2b, y2b = boxes[i]
bh, bw = y2b - y1b, x2b - x1b
yc, xc = int(random.uniform(0, h - bh)), int(random.uniform(0, w - bw)) # offset x, y
x1a, y1a, x2a, y2a = [xc, yc, xc + bw, yc + bh]
im[y1a:y2a, x1a:x2a] = im[y1b:y2b, x1b:x2b] # im4[ymin:ymax, xmin:xmax]
labels = np.append(labels, [[labels[i, 0], x1a, y1a, x2a, y2a]], axis=0)
return im, labels
def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32):
# Resize and pad image while meeting stride-multiple constraints
shape = im.shape[:2] # current shape [height, width]
if isinstance(new_shape, int):
new_shape = (new_shape, new_shape)
# Scale ratio (new / old)
r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
if not scaleup: # only scale down, do not scale up (for better val mAP)
r = min(r, 1.0)
# Compute padding
ratio = r, r # width, height ratios
new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1] # wh padding
if auto: # minimum rectangle
dw, dh = np.mod(dw, stride), np.mod(dh, stride) # wh padding
elif scaleFill: # stretch
dw, dh = 0.0, 0.0
new_unpad = (new_shape[1], new_shape[0])
ratio = new_shape[1] / shape[1], new_shape[0] / shape[0] # width, height ratios
dw /= 2 # divide padding into 2 sides
dh /= 2
if shape[::-1] != new_unpad: # resize
im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR)
top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
im = cv2.copyMakeBorder(im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color) # add border
return im, ratio, (dw, dh)
def random_perspective(im,
targets=(),
segments=(),
degrees=10,
translate=.1,
scale=.1,
shear=10,
perspective=0.0,
border=(0, 0)):
# torchvision.transforms.RandomAffine(degrees=(-10, 10), translate=(0.1, 0.1), scale=(0.9, 1.1), shear=(-10, 10))
# targets = [cls, xyxy]
height = im.shape[0] + border[0] * 2 # shape(h,w,c)
width = im.shape[1] + border[1] * 2
# Center
C = np.eye(3)
C[0, 2] = -im.shape[1] / 2 # x translation (pixels)
C[1, 2] = -im.shape[0] / 2 # y translation (pixels)
# Perspective
P = np.eye(3)
P[2, 0] = random.uniform(-perspective, perspective) # x perspective (about y)
P[2, 1] = random.uniform(-perspective, perspective) # y perspective (about x)
# Rotation and Scale
R = np.eye(3)
a = random.uniform(-degrees, degrees)
# a += random.choice([-180, -90, 0, 90]) # add 90deg rotations to small rotations
s = random.uniform(1 - scale, 1 + scale)
# s = 2 ** random.uniform(-scale, scale)
R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s)
# Shear
S = np.eye(3)
S[0, 1] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # x shear (deg)
S[1, 0] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # y shear (deg)
# Translation
T = np.eye(3)
T[0, 2] = random.uniform(0.5 - translate, 0.5 + translate) * width # x translation (pixels)
T[1, 2] = random.uniform(0.5 - translate, 0.5 + translate) * height # y translation (pixels)
# Combined rotation matrix
M = T @ S @ R @ P @ C # order of operations (right to left) is IMPORTANT
if (border[0] != 0) or (border[1] != 0) or (M != np.eye(3)).any(): # image changed
if perspective:
im = cv2.warpPerspective(im, M, dsize=(width, height), borderValue=(114, 114, 114))
else: # affine
im = cv2.warpAffine(im, M[:2], dsize=(width, height), borderValue=(114, 114, 114))
# Visualize
# import matplotlib.pyplot as plt
# ax = plt.subplots(1, 2, figsize=(12, 6))[1].ravel()
# ax[0].imshow(im[:, :, ::-1]) # base
# ax[1].imshow(im2[:, :, ::-1]) # warped
# Transform label coordinates
n = len(targets)
if n:
use_segments = any(x.any() for x in segments)
new = np.zeros((n, 4))
if use_segments: # warp segments
segments = resample_segments(segments) # upsample
for i, segment in enumerate(segments):
xy = np.ones((len(segment), 3))
xy[:, :2] = segment
xy = xy @ M.T # transform
xy = xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2] # perspective rescale or affine
# clip
new[i] = segment2box(xy, width, height)
else: # warp boxes
xy = np.ones((n * 4, 3))
xy[:, :2] = targets[:, [1, 2, 3, 4, 1, 4, 3, 2]].reshape(n * 4, 2) # x1y1, x2y2, x1y2, x2y1
xy = xy @ M.T # transform
xy = (xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2]).reshape(n, 8) # perspective rescale or affine
# create new boxes
x = xy[:, [0, 2, 4, 6]]
y = xy[:, [1, 3, 5, 7]]
new = np.concatenate((x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T
# clip
new[:, [0, 2]] = new[:, [0, 2]].clip(0, width)
new[:, [1, 3]] = new[:, [1, 3]].clip(0, height)
# filter candidates
i = box_candidates(box1=targets[:, 1:5].T * s, box2=new.T, area_thr=0.01 if use_segments else 0.10)
targets = targets[i]
targets[:, 1:5] = new[i]
return im, targets
def copy_paste(im, labels, segments, p=0.5):
# Implement Copy-Paste augmentation https://arxiv.org/abs/2012.07177, labels as nx5 np.array(cls, xyxy)
n = len(segments)
if p and n:
h, w, c = im.shape # height, width, channels
im_new = np.zeros(im.shape, np.uint8)
# calculate ioa first then select indexes randomly
boxes = np.stack([w - labels[:, 3], labels[:, 2], w - labels[:, 1], labels[:, 4]], axis=-1) # (n, 4)
ioa = bbox_ioa(boxes, labels[:, 1:5]) # intersection over area
indexes = np.nonzero((ioa < 0.30).all(1))[0] # (N, )
n = len(indexes)
for j in random.sample(list(indexes), k=round(p * n)):
l, box, s = labels[j], boxes[j], segments[j]
labels = np.concatenate((labels, [[l[0], *box]]), 0)
segments.append(np.concatenate((w - s[:, 0:1], s[:, 1:2]), 1))
cv2.drawContours(im_new, [segments[j].astype(np.int32)], -1, (1, 1, 1), cv2.FILLED)
result = cv2.flip(im, 1) # augment segments (flip left-right)
i = cv2.flip(im_new, 1).astype(bool)
im[i] = result[i] # cv2.imwrite('debug.jpg', im) # debug
return im, labels, segments
def cutout(im, labels, p=0.5):
# Applies image cutout augmentation https://arxiv.org/abs/1708.04552
if random.random() < p:
h, w = im.shape[:2]
scales = [0.5] * 1 + [0.25] * 2 + [0.125] * 4 + [0.0625] * 8 + [0.03125] * 16 # image size fraction
for s in scales:
mask_h = random.randint(1, int(h * s)) # create random masks
mask_w = random.randint(1, int(w * s))
# box
xmin = max(0, random.randint(0, w) - mask_w // 2)
ymin = max(0, random.randint(0, h) - mask_h // 2)
xmax = min(w, xmin + mask_w)
ymax = min(h, ymin + mask_h)
# apply random color mask
im[ymin:ymax, xmin:xmax] = [random.randint(64, 191) for _ in range(3)]
# return unobscured labels
if len(labels) and s > 0.03:
box = np.array([[xmin, ymin, xmax, ymax]], dtype=np.float32)
ioa = bbox_ioa(box, xywhn2xyxy(labels[:, 1:5], w, h))[0] # intersection over area
labels = labels[ioa < 0.60] # remove >60% obscured labels
return labels
def mixup(im, labels, im2, labels2):
# Applies MixUp augmentation https://arxiv.org/pdf/1710.09412.pdf
r = np.random.beta(32.0, 32.0) # mixup ratio, alpha=beta=32.0
im = (im * r + im2 * (1 - r)).astype(np.uint8)
labels = np.concatenate((labels, labels2), 0)
return im, labels
def box_candidates(box1, box2, wh_thr=2, ar_thr=100, area_thr=0.1, eps=1e-16): # box1(4,n), box2(4,n)
# Compute candidate boxes: box1 before augment, box2 after augment, wh_thr (pixels), aspect_ratio_thr, area_ratio
w1, h1 = box1[2] - box1[0], box1[3] - box1[1]
w2, h2 = box2[2] - box2[0], box2[3] - box2[1]
ar = np.maximum(w2 / (h2 + eps), h2 / (w2 + eps)) # aspect ratio
return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + eps) > area_thr) & (ar < ar_thr) # candidates
def classify_albumentations(
augment=True,
size=224,
scale=(0.08, 1.0),
ratio=(0.75, 1.0 / 0.75), # 0.75, 1.33
hflip=0.5,
vflip=0.0,
jitter=0.4,
mean=IMAGENET_MEAN,
std=IMAGENET_STD,
auto_aug=False):
# YOLOv5 classification Albumentations (optional, only used if package is installed)
prefix = colorstr('albumentations: ')
try:
import albumentations as A
from albumentations.pytorch import ToTensorV2
check_version(A.__version__, '1.0.3', hard=True) # version requirement
if augment: # Resize and crop
T = [A.RandomResizedCrop(height=size, width=size, scale=scale, ratio=ratio)]
if auto_aug:
# TODO: implement AugMix, AutoAug & RandAug in albumentation
LOGGER.info(f'{prefix}auto augmentations are currently not supported')
else:
if hflip > 0:
T += [A.HorizontalFlip(p=hflip)]
if vflip > 0:
T += [A.VerticalFlip(p=vflip)]
if jitter > 0:
color_jitter = (float(jitter),) * 3 # repeat value for brightness, contrast, satuaration, 0 hue
T += [A.ColorJitter(*color_jitter, 0)]
else: # Use fixed crop for eval set (reproducibility)
T = [A.SmallestMaxSize(max_size=size), A.CenterCrop(height=size, width=size)]
T += [A.Normalize(mean=mean, std=std), ToTensorV2()] # Normalize and convert to Tensor
LOGGER.info(prefix + ', '.join(f'{x}'.replace('always_apply=False, ', '') for x in T if x.p))
return A.Compose(T)
except ImportError: # package not installed, skip
LOGGER.warning(f'{prefix}⚠️ not found, install with `pip install albumentations` (recommended)')
except Exception as e:
LOGGER.info(f'{prefix}{e}')
def classify_transforms(size=224):
# Transforms to apply if albumentations not installed
assert isinstance(size, int), f'ERROR: classify_transforms size {size} must be integer, not (list, tuple)'
# T.Compose([T.ToTensor(), T.Resize(size), T.CenterCrop(size), T.Normalize(IMAGENET_MEAN, IMAGENET_STD)])
return T.Compose([CenterCrop(size), ToTensor(), T.Normalize(IMAGENET_MEAN, IMAGENET_STD)])
class LetterBox:
# YOLOv5 LetterBox class for image preprocessing, i.e. T.Compose([LetterBox(size), ToTensor()])
def __init__(self, size=(640, 640), auto=False, stride=32):
super().__init__()
self.h, self.w = (size, size) if isinstance(size, int) else size
self.auto = auto # pass max size integer, automatically solve for short side using stride
self.stride = stride # used with auto
def __call__(self, im): # im = np.array HWC
imh, imw = im.shape[:2]
r = min(self.h / imh, self.w / imw) # ratio of new/old
h, w = round(imh * r), round(imw * r) # resized image
hs, ws = (math.ceil(x / self.stride) * self.stride for x in (h, w)) if self.auto else self.h, self.w
top, left = round((hs - h) / 2 - 0.1), round((ws - w) / 2 - 0.1)
im_out = np.full((self.h, self.w, 3), 114, dtype=im.dtype)
im_out[top:top + h, left:left + w] = cv2.resize(im, (w, h), interpolation=cv2.INTER_LINEAR)
return im_out
class CenterCrop:
# YOLOv5 CenterCrop class for image preprocessing, i.e. T.Compose([CenterCrop(size), ToTensor()])
def __init__(self, size=640):
super().__init__()
self.h, self.w = (size, size) if isinstance(size, int) else size
def __call__(self, im): # im = np.array HWC
imh, imw = im.shape[:2]
m = min(imh, imw) # min dimension
top, left = (imh - m) // 2, (imw - m) // 2
return cv2.resize(im[top:top + m, left:left + m], (self.w, self.h), interpolation=cv2.INTER_LINEAR)
class ToTensor:
# YOLOv5 ToTensor class for image preprocessing, i.e. T.Compose([LetterBox(size), ToTensor()])
def __init__(self, half=False):
super().__init__()
self.half = half
def __call__(self, im): # im = np.array HWC in BGR order
im = np.ascontiguousarray(im.transpose((2, 0, 1))[::-1]) # HWC to CHW -> BGR to RGB -> contiguous
im = torch.from_numpy(im) # to torch
im = im.half() if self.half else im.float() # uint8 to fp16/32
im /= 255.0 # 0-255 to 0.0-1.0
return im
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment