Skip to content

GitLab

"vscode:/vscode.git/clone" did not exist on "6425d46cfc15b5a7073a30cee4024434eeeac7a0"
Commit 6d71b439 authored by zhangwenwei's avatar zhangwenwei
Browse files

Refactor optimizer and samplers

parent ba492be7
Show whitespace changes
Inline Side-by-side
Showing with 347 additions and 762 deletions
mmdet3d/core/bbox/samplers/combined_sampler.py deleted 100644 → 0
View file @ ba492be7
from ..assign_sampling import build_sampler
from .base_sampler import BaseSampler
class CombinedSampler(BaseSampler):
def __init__(self, pos_sampler, neg_sampler, **kwargs):
super(CombinedSampler, self).__init__(**kwargs)
self.pos_sampler = build_sampler(pos_sampler, **kwargs)
self.neg_sampler = build_sampler(neg_sampler, **kwargs)
def _sample_pos(self, **kwargs):
raise NotImplementedError
def _sample_neg(self, **kwargs):
raise NotImplementedError
mmdet3d/core/bbox/samplers/instance_balanced_pos_sampler.py deleted 100644 → 0
View file @ ba492be7
import numpy as np
import torch
from .random_sampler import RandomSampler
class InstanceBalancedPosSampler(RandomSampler):
def _sample_pos(self, assign_result, num_expected, **kwargs):
pos_inds = torch.nonzero(assign_result.gt_inds > 0)
if pos_inds.numel() != 0:
pos_inds = pos_inds.squeeze(1)
if pos_inds.numel() <= num_expected:
return pos_inds
else:
unique_gt_inds = assign_result.gt_inds[pos_inds].unique()
num_gts = len(unique_gt_inds)
num_per_gt = int(round(num_expected / float(num_gts)) + 1)
sampled_inds = []
for i in unique_gt_inds:
inds = torch.nonzero(assign_result.gt_inds == i.item())
if inds.numel() != 0:
inds = inds.squeeze(1)
else:
continue
if len(inds) > num_per_gt:
inds = self.random_choice(inds, num_per_gt)
sampled_inds.append(inds)
sampled_inds = torch.cat(sampled_inds)
if len(sampled_inds) < num_expected:
num_extra = num_expected - len(sampled_inds)
extra_inds = np.array(
list(set(pos_inds.cpu()) - set(sampled_inds.cpu())))
if len(extra_inds) > num_extra:
extra_inds = self.random_choice(extra_inds, num_extra)
extra_inds = torch.from_numpy(extra_inds).to(
assign_result.gt_inds.device).long()
sampled_inds = torch.cat([sampled_inds, extra_inds])
elif len(sampled_inds) > num_expected:
sampled_inds = self.random_choice(sampled_inds, num_expected)
return sampled_inds
mmdet3d/core/bbox/samplers/iou_balanced_neg_sampler.py deleted 100644 → 0
View file @ ba492be7
import numpy as np
import torch
from .random_sampler import RandomSampler
class IoUBalancedNegSampler(RandomSampler):
"""IoU Balanced Sampling
arXiv: https://arxiv.org/pdf/1904.02701.pdf (CVPR 2019)
Sampling proposals according to their IoU. `floor_fraction` of needed RoIs
are sampled from proposals whose IoU are lower than `floor_thr` randomly.
The others are sampled from proposals whose IoU are higher than
`floor_thr`. These proposals are sampled from some bins evenly, which are
split by `num_bins` via IoU evenly.
Args:
num (int): number of proposals.
pos_fraction (float): fraction of positive proposals.
floor_thr (float): threshold (minimum) IoU for IoU balanced sampling,
set to -1 if all using IoU balanced sampling.
floor_fraction (float): sampling fraction of proposals under floor_thr.
num_bins (int): number of bins in IoU balanced sampling.
"""
def __init__(self,
num,
pos_fraction,
floor_thr=-1,
floor_fraction=0,
num_bins=3,
**kwargs):
super(IoUBalancedNegSampler, self).__init__(num, pos_fraction,
**kwargs)
assert floor_thr >= 0 or floor_thr == -1
assert 0 <= floor_fraction <= 1
assert num_bins >= 1
self.floor_thr = floor_thr
self.floor_fraction = floor_fraction
self.num_bins = num_bins
def sample_via_interval(self, max_overlaps, full_set, num_expected):
max_iou = max_overlaps.max()
iou_interval = (max_iou - self.floor_thr) / self.num_bins
per_num_expected = int(num_expected / self.num_bins)
sampled_inds = []
for i in range(self.num_bins):
start_iou = self.floor_thr + i * iou_interval
end_iou = self.floor_thr + (i + 1) * iou_interval
tmp_set = set(
np.where(
np.logical_and(max_overlaps >= start_iou,
max_overlaps < end_iou))[0])
tmp_inds = list(tmp_set & full_set)
if len(tmp_inds) > per_num_expected:
tmp_sampled_set = self.random_choice(tmp_inds,
per_num_expected)
else:
tmp_sampled_set = np.array(tmp_inds, dtype=np.int)
sampled_inds.append(tmp_sampled_set)
sampled_inds = np.concatenate(sampled_inds)
if len(sampled_inds) < num_expected:
num_extra = num_expected - len(sampled_inds)
extra_inds = np.array(list(full_set - set(sampled_inds)))
if len(extra_inds) > num_extra:
extra_inds = self.random_choice(extra_inds, num_extra)
sampled_inds = np.concatenate([sampled_inds, extra_inds])
return sampled_inds
def _sample_neg(self, assign_result, num_expected, **kwargs):
neg_inds = torch.nonzero(assign_result.gt_inds == 0)
if neg_inds.numel() != 0:
neg_inds = neg_inds.squeeze(1)
if len(neg_inds) <= num_expected:
return neg_inds
else:
max_overlaps = assign_result.max_overlaps.cpu().numpy()
# balance sampling for negative samples
neg_set = set(neg_inds.cpu().numpy())
if self.floor_thr > 0:
floor_set = set(
np.where(
np.logical_and(max_overlaps >= 0,
max_overlaps < self.floor_thr))[0])
iou_sampling_set = set(
np.where(max_overlaps >= self.floor_thr)[0])
elif self.floor_thr == 0:
floor_set = set(np.where(max_overlaps == 0)[0])
iou_sampling_set = set(
np.where(max_overlaps > self.floor_thr)[0])
else:
floor_set = set()
iou_sampling_set = set(
np.where(max_overlaps > self.floor_thr)[0])
floor_neg_inds = list(floor_set & neg_set)
iou_sampling_neg_inds = list(iou_sampling_set & neg_set)
num_expected_iou_sampling = int(num_expected *
(1 - self.floor_fraction))
if len(iou_sampling_neg_inds) > num_expected_iou_sampling:
if self.num_bins >= 2:
iou_sampled_inds = self.sample_via_interval(
max_overlaps, set(iou_sampling_neg_inds),
num_expected_iou_sampling)
else:
iou_sampled_inds = self.random_choice(
iou_sampling_neg_inds, num_expected_iou_sampling)
else:
iou_sampled_inds = np.array(
iou_sampling_neg_inds, dtype=np.int)
num_expected_floor = num_expected - len(iou_sampled_inds)
if len(floor_neg_inds) > num_expected_floor:
sampled_floor_inds = self.random_choice(
floor_neg_inds, num_expected_floor)
else:
sampled_floor_inds = np.array(floor_neg_inds, dtype=np.int)
sampled_inds = np.concatenate(
(sampled_floor_inds, iou_sampled_inds))
if len(sampled_inds) < num_expected:
num_extra = num_expected - len(sampled_inds)
extra_inds = np.array(list(neg_set - set(sampled_inds)))
if len(extra_inds) > num_extra:
extra_inds = self.random_choice(extra_inds, num_extra)
sampled_inds = np.concatenate((sampled_inds, extra_inds))
sampled_inds = torch.from_numpy(sampled_inds).long().to(
assign_result.gt_inds.device)
return sampled_inds
mmdet3d/core/bbox/samplers/ohem_sampler.py deleted 100644 → 0
View file @ ba492be7
import torch
from ..transforms import bbox2roi
from .base_sampler import BaseSampler
class OHEMSampler(BaseSampler):
def __init__(self,
num,
pos_fraction,
context,
neg_pos_ub=-1,
add_gt_as_proposals=True,
**kwargs):
super(OHEMSampler, self).__init__(num, pos_fraction, neg_pos_ub,
add_gt_as_proposals)
if not hasattr(context, 'num_stages'):
self.bbox_roi_extractor = context.bbox_roi_extractor
self.bbox_head = context.bbox_head
else:
self.bbox_roi_extractor = context.bbox_roi_extractor[
context.current_stage]
self.bbox_head = context.bbox_head[context.current_stage]
def hard_mining(self, inds, num_expected, bboxes, labels, feats):
with torch.no_grad():
rois = bbox2roi([bboxes])
bbox_feats = self.bbox_roi_extractor(
feats[:self.bbox_roi_extractor.num_inputs], rois)
cls_score, _ = self.bbox_head(bbox_feats)
loss = self.bbox_head.loss(
cls_score=cls_score,
bbox_pred=None,
labels=labels,
label_weights=cls_score.new_ones(cls_score.size(0)),
bbox_targets=None,
bbox_weights=None,
reduction_override='none')['loss_cls']
_, topk_loss_inds = loss.topk(num_expected)
return inds[topk_loss_inds]
def _sample_pos(self,
assign_result,
num_expected,
bboxes=None,
feats=None,
**kwargs):
# Sample some hard positive samples
pos_inds = torch.nonzero(assign_result.gt_inds > 0)
if pos_inds.numel() != 0:
pos_inds = pos_inds.squeeze(1)
if pos_inds.numel() <= num_expected:
return pos_inds
else:
return self.hard_mining(pos_inds, num_expected, bboxes[pos_inds],
assign_result.labels[pos_inds], feats)
def _sample_neg(self,
assign_result,
num_expected,
bboxes=None,
feats=None,
**kwargs):
# Sample some hard negative samples
neg_inds = torch.nonzero(assign_result.gt_inds == 0)
if neg_inds.numel() != 0:
neg_inds = neg_inds.squeeze(1)
if len(neg_inds) <= num_expected:
return neg_inds
else:
return self.hard_mining(neg_inds, num_expected, bboxes[neg_inds],
assign_result.labels[neg_inds], feats)
mmdet3d/core/bbox/samplers/pseudo_sampler.py deleted 100644 → 0
View file @ ba492be7
import torch
from .base_sampler import BaseSampler
from .sampling_result import SamplingResult
class PseudoSampler(BaseSampler):
def __init__(self, **kwargs):
pass
def _sample_pos(self, **kwargs):
raise NotImplementedError
def _sample_neg(self, **kwargs):
raise NotImplementedError
def sample(self, assign_result, bboxes, gt_bboxes, **kwargs):
pos_inds = torch.nonzero(
assign_result.gt_inds > 0).squeeze(-1).unique()
neg_inds = torch.nonzero(
assign_result.gt_inds == 0).squeeze(-1).unique()
gt_flags = bboxes.new_zeros(bboxes.shape[0], dtype=torch.uint8)
sampling_result = SamplingResult(pos_inds, neg_inds, bboxes, gt_bboxes,
assign_result, gt_flags)
return sampling_result
mmdet3d/core/bbox/samplers/random_sampler.py deleted 100644 → 0
View file @ ba492be7
import numpy as np
import torch
from .base_sampler import BaseSampler
class RandomSampler(BaseSampler):
def __init__(self,
num,
pos_fraction,
neg_pos_ub=-1,
add_gt_as_proposals=True,
**kwargs):
super(RandomSampler, self).__init__(num, pos_fraction, neg_pos_ub,
add_gt_as_proposals)
@staticmethod
def random_choice(gallery, num):
"""Random select some elements from the gallery.
It seems that Pytorch's implementation is slower than numpy so we use
numpy to randperm the indices.
"""
assert len(gallery) >= num
if isinstance(gallery, list):
gallery = np.array(gallery)
cands = np.arange(len(gallery))
np.random.shuffle(cands)
rand_inds = cands[:num]
if not isinstance(gallery, np.ndarray):
rand_inds = torch.from_numpy(rand_inds).long().to(gallery.device)
return gallery[rand_inds]
def _sample_pos(self, assign_result, num_expected, **kwargs):
"""Randomly sample some positive samples."""
pos_inds = torch.nonzero(assign_result.gt_inds > 0)
if pos_inds.numel() != 0:
pos_inds = pos_inds.squeeze(1)
if pos_inds.numel() <= num_expected:
return pos_inds
else:
return self.random_choice(pos_inds, num_expected)
def _sample_neg(self, assign_result, num_expected, **kwargs):
"""Randomly sample some negative samples."""
neg_inds = torch.nonzero(assign_result.gt_inds == 0)
if neg_inds.numel() != 0:
neg_inds = neg_inds.squeeze(1)
if len(neg_inds) <= num_expected:
return neg_inds
else:
return self.random_choice(neg_inds, num_expected)
mmdet3d/core/bbox/samplers/sampling_result.py deleted 100644 → 0
View file @ ba492be7
import torch
class SamplingResult(object):
def __init__(self, pos_inds, neg_inds, bboxes, gt_bboxes, assign_result,
gt_flags):
self.pos_inds = pos_inds
self.neg_inds = neg_inds
self.pos_bboxes = bboxes[pos_inds]
self.neg_bboxes = bboxes[neg_inds]
self.pos_is_gt = gt_flags[pos_inds]
self.num_gts = gt_bboxes.shape[0]
self.pos_assigned_gt_inds = assign_result.gt_inds[pos_inds] - 1
self.pos_gt_bboxes = gt_bboxes[self.pos_assigned_gt_inds, :]
if assign_result.labels is not None:
self.pos_gt_labels = assign_result.labels[pos_inds]
else:
self.pos_gt_labels = None
@property
def bboxes(self):
return torch.cat([self.pos_bboxes, self.neg_bboxes])
mmdet3d/core/bbox/transforms.py
View file @ 6d71b439
import mmcv
import numpy as np
import torch import torch
def bbox2delta(proposals, gt, means=[0, 0, 0, 0], stds=[1, 1, 1, 1]):
assert proposals.size() == gt.size()
proposals = proposals.float()
gt = gt.float()
px = (proposals[..., 0] + proposals[..., 2]) * 0.5
py = (proposals[..., 1] + proposals[..., 3]) * 0.5
pw = proposals[..., 2] - proposals[..., 0]
ph = proposals[..., 3] - proposals[..., 1]
gx = (gt[..., 0] + gt[..., 2]) * 0.5
gy = (gt[..., 1] + gt[..., 3]) * 0.5
gw = gt[..., 2] - gt[..., 0]
gh = gt[..., 3] - gt[..., 1]
dx = (gx - px) / pw
dy = (gy - py) / ph
dw = torch.log(gw / pw)
dh = torch.log(gh / ph)
deltas = torch.stack([dx, dy, dw, dh], dim=-1)
means = deltas.new_tensor(means).unsqueeze(0)
stds = deltas.new_tensor(stds).unsqueeze(0)
deltas = deltas.sub_(means).div_(stds)
return deltas
def delta2bbox(rois,
deltas,
means=[0, 0, 0, 0],
stds=[1, 1, 1, 1],
max_shape=None,
wh_ratio_clip=16 / 1000):
"""
Apply deltas to shift/scale base boxes.
Typically the rois are anchor or proposed bounding boxes and the deltas are
network outputs used to shift/scale those boxes.
Args:
rois (Tensor): boxes to be transformed. Has shape (N, 4)
deltas (Tensor): encoded offsets with respect to each roi.
Has shape (N, 4). Note N = num_anchors * W * H when rois is a grid
of anchors. Offset encoding follows [1]_.
means (list): denormalizing means for delta coordinates
stds (list): denormalizing standard deviation for delta coordinates
max_shape (tuple[int, int]): maximum bounds for boxes. specifies (H, W)
wh_ratio_clip (float): maximum aspect ratio for boxes.
Returns:
Tensor: boxes with shape (N, 4), where columns represent
tl_x, tl_y, br_x, br_y.
References:
.. [1] https://arxiv.org/abs/1311.2524
Example:
>>> rois = torch.Tensor([[ 0., 0., 1., 1.],
>>> [ 0., 0., 1., 1.],
>>> [ 0., 0., 1., 1.],
>>> [ 5., 5., 5., 5.]])
>>> deltas = torch.Tensor([[ 0., 0., 0., 0.],
>>> [ 1., 1., 1., 1.],
>>> [ 0., 0., 2., -1.],
>>> [ 0.7, -1.9, -0.5, 0.3]])
>>> delta2bbox(rois, deltas, max_shape=(32, 32))
tensor([[0.0000, 0.0000, 1.0000, 1.0000],
[0.2817, 0.2817, 4.7183, 4.7183],
[0.0000, 0.6321, 7.3891, 0.3679],
[5.8967, 2.9251, 5.5033, 3.2749]])
"""
means = deltas.new_tensor(means).repeat(1, deltas.size(1) // 4)
stds = deltas.new_tensor(stds).repeat(1, deltas.size(1) // 4)
denorm_deltas = deltas * stds + means
dx = denorm_deltas[:, 0::4]
dy = denorm_deltas[:, 1::4]
dw = denorm_deltas[:, 2::4]
dh = denorm_deltas[:, 3::4]
max_ratio = np.abs(np.log(wh_ratio_clip))
dw = dw.clamp(min=-max_ratio, max=max_ratio)
dh = dh.clamp(min=-max_ratio, max=max_ratio)
# Compute center of each roi
px = ((rois[:, 0] + rois[:, 2]) * 0.5).unsqueeze(1).expand_as(dx)
py = ((rois[:, 1] + rois[:, 3]) * 0.5).unsqueeze(1).expand_as(dy)
# Compute width/height of each roi
pw = (rois[:, 2] - rois[:, 0]).unsqueeze(1).expand_as(dw)
ph = (rois[:, 3] - rois[:, 1]).unsqueeze(1).expand_as(dh)
# Use exp(network energy) to enlarge/shrink each roi
gw = pw * dw.exp()
gh = ph * dh.exp()
# Use network energy to shift the center of each roi
gx = torch.addcmul(px, 1, pw, dx) # gx = px + pw * dx
gy = torch.addcmul(py, 1, ph, dy) # gy = py + ph * dy
# Convert center-xy/width/height to top-left, bottom-right
x1 = gx - gw * 0.5
y1 = gy - gh * 0.5
x2 = gx + gw * 0.5
y2 = gy + gh * 0.5
if max_shape is not None:
x1 = x1.clamp(min=0, max=max_shape[1])
y1 = y1.clamp(min=0, max=max_shape[0])
x2 = x2.clamp(min=0, max=max_shape[1])
y2 = y2.clamp(min=0, max=max_shape[0])
bboxes = torch.stack([x1, y1, x2, y2], dim=-1).view_as(deltas)
return bboxes
def bbox_flip(bboxes, img_shape):
"""Flip bboxes horizontally.
Args:
bboxes(Tensor or ndarray): Shape (..., 4*k)
img_shape(tuple): Image shape.
Returns:
Same type as `bboxes`: Flipped bboxes.
"""
if isinstance(bboxes, torch.Tensor):
assert bboxes.shape[-1] % 4 == 0
flipped = bboxes.clone()
flipped[:, 0::4] = img_shape[1] - bboxes[:, 2::4]
flipped[:, 2::4] = img_shape[1] - bboxes[:, 0::4]
return flipped
elif isinstance(bboxes, np.ndarray):
return mmcv.bbox_flip(bboxes, img_shape)
def bbox_mapping(bboxes, img_shape, scale_factor, flip):
"""Map bboxes from the original image scale to testing scale"""
new_bboxes = bboxes * scale_factor
if flip:
new_bboxes = bbox_flip(new_bboxes, img_shape)
return new_bboxes
def bbox_mapping_back(bboxes, img_shape, scale_factor, flip):
"""Map bboxes from testing scale to original image scale"""
new_bboxes = bbox_flip(bboxes, img_shape) if flip else bboxes
new_bboxes = new_bboxes / scale_factor
return new_bboxes
def bbox2roi(bbox_list):
"""Convert a list of bboxes to roi format.
Args:
bbox_list (list[Tensor]): a list of bboxes corresponding to a batch
of images.
Returns:
Tensor: shape (n, 5), [batch_ind, x1, y1, x2, y2]
"""
rois_list = []
for img_id, bboxes in enumerate(bbox_list):
if bboxes.size(0) > 0:
img_inds = bboxes.new_full((bboxes.size(0), 1), img_id)
rois = torch.cat([img_inds, bboxes[:, :4]], dim=-1)
else:
rois = bboxes.new_zeros((0, 5))
rois_list.append(rois)
rois = torch.cat(rois_list, 0)
return rois
def roi2bbox(rois):
bbox_list = []
img_ids = torch.unique(rois[:, 0].cpu(), sorted=True)
for img_id in img_ids:
inds = (rois[:, 0] == img_id.item())
bbox = rois[inds, 1:]
bbox_list.append(bbox)
return bbox_list
def bbox2result_coco(bboxes, labels, num_classes):
"""Convert detection results to a list of numpy arrays.
Args:
bboxes (Tensor): shape (n, 5)
labels (Tensor): shape (n, )
num_classes (int): class number, including background class
Returns:
list(ndarray): bbox results of each class
"""
if bboxes.shape[0] == 0:
return [np.zeros((0, 5), dtype=np.float32) for i in range(num_classes)]
else:
bboxes = bboxes.cpu().numpy()
labels = labels.cpu().numpy()
return [bboxes[labels == i, :] for i in range(num_classes)]
def distance2bbox(points, distance, max_shape=None):
"""Decode distance prediction to bounding box.
Args:
points (Tensor): Shape (n, 2), [x, y].
distance (Tensor): Distance from the given point to 4
boundaries (left, top, right, bottom).
max_shape (tuple): Shape of the image.
Returns:
Tensor: Decoded bboxes.
"""
x1 = points[:, 0] - distance[:, 0]
y1 = points[:, 1] - distance[:, 1]
x2 = points[:, 0] + distance[:, 2]
y2 = points[:, 1] + distance[:, 3]
if max_shape is not None:
x1 = x1.clamp(min=0, max=max_shape[1])
y1 = y1.clamp(min=0, max=max_shape[0])
x2 = x2.clamp(min=0, max=max_shape[1])
y2 = y2.clamp(min=0, max=max_shape[0])
return torch.stack([x1, y1, x2, y2], -1)
def transform_lidar_to_cam(boxes_lidar): def transform_lidar_to_cam(boxes_lidar):
""" """
Only transform format, not exactly in camera coords Only transform format, not exactly in camera coords
......
mmdet3d/core/optimizer/__init__.py
View file @ 6d71b439
from .builder import build_optimizer from .cocktail_constructor import CocktailOptimizerConstructor
from .mix_optimizer import MixedOptimizer from .cocktail_optimizer import CocktailOptimizer
from .registry import OPTIMIZERS
__all__ = ['OPTIMIZERS', 'build_optimizer', 'MixedOptimizer'] __all__ = ['CocktailOptimizerConstructor', 'CocktailOptimizer']
mmdet3d/core/optimizer/builder.py deleted 100644 → 0
View file @ ba492be7
import re
import torch
from mmdet.utils import build_from_cfg, get_root_logger
from .registry import OPTIMIZERS
def build_optimizer(model, optimizer_cfg):
"""Build optimizer from configs.
Args:
model (:obj:`nn.Module`): The model with parameters to be optimized.
optimizer_cfg (dict): The config dict of the optimizer.
Positional fields are:
- type: class name of the optimizer.
- lr: base learning rate.
Optional fields are:
- any arguments of the corresponding optimizer type, e.g.,
weight_decay, momentum, etc.
- paramwise_options: a dict with 4 accepted fileds
(bias_lr_mult, bias_decay_mult, norm_decay_mult,
dwconv_decay_mult).
`bias_lr_mult` and `bias_decay_mult` will be multiplied to
the lr and weight decay respectively for all bias parameters
(except for the normalization layers), and
`norm_decay_mult` will be multiplied to the weight decay
for all weight and bias parameters of normalization layers.
`dwconv_decay_mult` will be multiplied to the weight decay
for all weight and bias parameters of depthwise conv layers.
Returns:
torch.optim.Optimizer: The initialized optimizer.
Example:
>>> import torch
>>> model = torch.nn.modules.Conv1d(1, 1, 1)
>>> optimizer_cfg = dict(type='SGD', lr=0.01, momentum=0.9,
>>> weight_decay=0.0001)
>>> optimizer = build_optimizer(model, optimizer_cfg)
"""
if hasattr(model, 'module'):
model = model.module
optimizer_cfg = optimizer_cfg.copy()
if isinstance(optimizer_cfg, list):
# Assume paramwise_options is None if optimizer_cfg is list
from .mix_optimizer import MixedOptimizer
logger = get_root_logger()
keys = [optimizer.pop('key') for optimizer in optimizer_cfg]
keys_params = {key: [] for key in keys}
keys_params_name = {key: [] for key in keys}
keys_optimizer = []
for name, param in model.named_parameters():
param_group = {'params': [param]}
find_flag = False
for key in keys:
if key in name:
keys_params[key].append(param_group)
keys_params_name[key].append(name)
find_flag = True
break
assert find_flag, 'key {} is not matched to any optimizer'.format(
name)
step_intervals = []
for key, single_cfg in zip(keys, optimizer_cfg):
optimizer_cls = getattr(torch.optim, single_cfg.pop('type'))
step_intervals.append(single_cfg.pop('step_interval', 1))
single_optim = optimizer_cls(keys_params[key], **single_cfg)
keys_optimizer.append(single_optim)
logger.info('{} optimizes key:\n {}\n'.format(
optimizer_cls.__name__, keys_params_name[key]))
mix_optimizer = MixedOptimizer(keys_optimizer, step_intervals)
return mix_optimizer
else:
paramwise_options = optimizer_cfg.pop('paramwise_options', None)
# if no paramwise option is specified, just use the global setting
if paramwise_options is None:
params = model.parameters()
else:
assert isinstance(paramwise_options, dict)
# get base lr and weight decay
base_lr = optimizer_cfg['lr']
base_wd = optimizer_cfg.get('weight_decay', None)
# weight_decay must be explicitly specified if mult is specified
if ('bias_decay_mult' in paramwise_options
or 'norm_decay_mult' in paramwise_options
or 'dwconv_decay_mult' in paramwise_options):
assert base_wd is not None
# get param-wise options
bias_lr_mult = paramwise_options.get('bias_lr_mult', 1.)
bias_decay_mult = paramwise_options.get('bias_decay_mult', 1.)
norm_decay_mult = paramwise_options.get('norm_decay_mult', 1.)
dwconv_decay_mult = paramwise_options.get('dwconv_decay_mult', 1.)
named_modules = dict(model.named_modules())
# set param-wise lr and weight decay
params = []
for name, param in model.named_parameters():
param_group = {'params': [param]}
if not param.requires_grad:
# FP16 training needs to copy gradient/weight between master
# weight copy and model weight, it is convenient to keep all
# parameters here to align with model.parameters()
params.append(param_group)
continue
# for norm layers, overwrite the weight decay of weight and bias
# TODO: obtain the norm layer prefixes dynamically
if re.search(r'(bn|gn)(\d+)?.(weight|bias)', name):
if base_wd is not None:
param_group['weight_decay'] = base_wd * norm_decay_mult
# for other layers, overwrite both lr and weight decay of bias
elif name.endswith('.bias'):
param_group['lr'] = base_lr * bias_lr_mult
if base_wd is not None:
param_group['weight_decay'] = base_wd * bias_decay_mult
module_name = name.replace('.weight', '').replace('.bias', '')
if module_name in named_modules and base_wd is not None:
module = named_modules[module_name]
# if this Conv2d is depthwise Conv2d
if isinstance(module, torch.nn.Conv2d) and \
module.in_channels == module.groups:
param_group['weight_decay'] = base_wd * dwconv_decay_mult
# otherwise use the global settings
params.append(param_group)
optimizer_cfg['params'] = params
return build_from_cfg(optimizer_cfg, OPTIMIZERS)
mmdet3d/core/optimizer/cocktail_constructor.py 0 → 100644
View file @ 6d71b439
from mmcv.utils import build_from_cfg
from mmdet.core.optimizer import OPTIMIZER_BUILDERS, OPTIMIZERS
from mmdet.utils import get_root_logger
from .cocktail_optimizer import CocktailOptimizer
@OPTIMIZER_BUILDERS.register_module
class CocktailOptimizerConstructor(object):
"""Default constructor for optimizers.
Attributes:
model (:obj:`nn.Module`): The model with parameters to be optimized.
optimizer_cfg (dict): The config dict of the optimizer.
Positional fields are:
- type: class name of the optimizer.
- lr: base learning rate.
Optional fields are:
- any arguments of the corresponding optimizer type, e.g.,
weight_decay, momentum, etc.
paramwise_cfg (dict, optional): Parameter-wise options. Accepted fields
are:
- bias_lr_mult: It will be multiplied to the learning rate for
all bias parameters (except for those in normalization layers).
- bias_decay_mult: It will be multiplied to the weight decay for
all bias parameters (except for those in normalization layers and
depthwise conv layers).
- norm_decay_mult: will be multiplied to the weight decay
for all weight and bias parameters of normalization layers.
- dwconv_decay_mult: will be multiplied to the weight decay
for all weight and bias parameters of depthwise conv layers.
Example:
>>> import torch
>>> import torch.nn as nn
>>> model = nn.ModuleDict({
>>> 'pts': nn.modules.Conv1d(1, 1, 1, bias=False),
>>> 'img': nn.modules.Conv1d(1, 1, 1, bias=False)
>>> })
>>> optimizer_cfg = dict(
>>> pts=dict(type='AdamW', lr=0.001,
>>> weight_decay=0.01, step_interval=1),
>>> img=dict(type='SGD', lr=0.02, momentum=0.9,
>>> weight_decay=0.0001, step_interval=2))
>>> optim_builder = CocktailOptimizerConstructor(optimizer_cfg)
>>> optimizer = optim_builder(model)
>>> print(optimizer)
CocktailOptimizer (
Update interval: 1
AdamW (
Parameter Group 0
amsgrad: False
betas: (0.9, 0.999)
eps: 1e-08
lr: 0.001
weight_decay: 0.01
),
Update interval: 2
SGD (
Parameter Group 0
dampening: 0
lr: 0.02
momentum: 0.9
nesterov: False
weight_decay: 0.0001
),
)
"""
def __init__(self, optimizer_cfg, paramwise_cfg=None):
if not isinstance(optimizer_cfg, dict):
raise TypeError('optimizer_cfg should be a dict',
'but got {}'.format(type(optimizer_cfg)))
assert paramwise_cfg is None, \
'Parameter wise config is not supported in Cocktail Optimizer'
self.optimizer_cfg = optimizer_cfg
def __call__(self, model):
if hasattr(model, 'module'):
model = model.module
optimizer_cfg = self.optimizer_cfg.copy()
logger = get_root_logger()
keys_prefix = [key_prefix for key_prefix in optimizer_cfg.keys()]
keys_params = {key: [] for key in keys_prefix}
keys_params_name = {key: [] for key in keys_prefix}
keys_optimizer = []
for name, param in model.named_parameters():
param_group = {'params': [param]}
find_flag = False
for key in keys_prefix:
if key in name:
keys_params[key].append(param_group)
keys_params_name[key].append(name)
find_flag = True
break
assert find_flag, 'key {} is not matched to any optimizer'.format(
name)
step_intervals = []
for key, single_cfg in optimizer_cfg.items():
step_intervals.append(single_cfg.pop('step_interval', 1))
single_cfg['params'] = keys_params[key]
single_optim = build_from_cfg(single_cfg, OPTIMIZERS)
keys_optimizer.append(single_optim)
logger.info('{} optimizes key:\n {}\n'.format(
single_cfg['type'], keys_params_name[key]))
cocktail_optimizer = CocktailOptimizer(keys_optimizer, step_intervals)
return cocktail_optimizer
mmdet3d/core/optimizer/mix_optimizer.py → mmdet3d/core/optimizer/cocktail_optimizer.py
View file @ 6d71b439
from torch.optim import Optimizer from torch.optim import Optimizer
from .registry import OPTIMIZERS from mmdet.core.optimizer import OPTIMIZERS
@OPTIMIZERS.register_module @OPTIMIZERS.register_module
class MixedOptimizer(Optimizer): class CocktailOptimizer(Optimizer):
"""Mixed Optimizer that contains multiple optimizers """Cocktail Optimizer that contains multiple optimizers
This optimizer applies the cocktail optimzation for multi-modality models. This optimizer applies the cocktail optimzation for multi-modality models.
...@@ -36,8 +36,9 @@ class MixedOptimizer(Optimizer): ...@@ -36,8 +36,9 @@ class MixedOptimizer(Optimizer):
def __repr__(self): def __repr__(self):
format_string = self.__class__.__name__ + ' (\n' format_string = self.__class__.__name__ + ' (\n'
for optimizer in self.optimizers: for optimizer, interval in zip(self.optimizers, self.step_intervals):
format_string += '\t' + optimizer.__repr__ + ',\n' format_string += 'Update interval: {}\n'.format(interval)
format_string += optimizer.__repr__().replace('\n', '\n ') + ',\n'
format_string += ')' format_string += ')'
return format_string return format_string
......
mmdet3d/core/optimizer/registry.py deleted 100644 → 0
View file @ ba492be7
import inspect
import torch
from mmdet.utils import Registry
OPTIMIZERS = Registry('optimizer')
def register_torch_optimizers():
torch_optimizers = []
for module_name in dir(torch.optim):
if module_name.startswith('__'):
continue
_optim = getattr(torch.optim, module_name)
if inspect.isclass(_optim) and issubclass(_optim,
torch.optim.Optimizer):
OPTIMIZERS.register_module(_optim)
torch_optimizers.append(module_name)
return torch_optimizers
TORCH_OPTIMIZERS = register_torch_optimizers()
mmdet3d/models/anchor_heads/boxvelo_head.py
View file @ 6d71b439
...@@ -49,7 +49,7 @@ class Anchor3DVeloHead(SECONDHead): ...@@ -49,7 +49,7 @@ class Anchor3DVeloHead(SECONDHead):
dir_limit_offset=1, dir_limit_offset=1,
target_means=(.0, .0, .0, .0), target_means=(.0, .0, .0, .0),
target_stds=(1.0, 1.0, 1.0, 1.0), target_stds=(1.0, 1.0, 1.0, 1.0),
bbox_coder=dict(type='ResidualCoder', ), bbox_coder=dict(type='Residual3DBoxCoder', ),
loss_cls=dict( loss_cls=dict(
type='CrossEntropyLoss', type='CrossEntropyLoss',
use_sigmoid=True, use_sigmoid=True,
......
mmdet3d/models/anchor_heads/second_head.py
View file @ 6d71b439
...@@ -55,7 +55,7 @@ class SECONDHead(nn.Module, AnchorTrainMixin): ...@@ -55,7 +55,7 @@ class SECONDHead(nn.Module, AnchorTrainMixin):
dir_limit_offset=1, dir_limit_offset=1,
target_means=(.0, .0, .0, .0), target_means=(.0, .0, .0, .0),
target_stds=(1.0, 1.0, 1.0, 1.0), target_stds=(1.0, 1.0, 1.0, 1.0),
bbox_coder=dict(type='ResidualCoder'), bbox_coder=dict(type='Residual3DBoxCoder'),
loss_cls=dict( loss_cls=dict(
type='CrossEntropyLoss', type='CrossEntropyLoss',
use_sigmoid=True, use_sigmoid=True,
......
tests/test_forward.py 0 → 100644
View file @ 6d71b439
"""
Test model forward process
CommandLine:
pytest tests/test_forward.py
xdoctest tests/test_forward.py zero
"""
import copy
from os.path import dirname, exists, join
import numpy as np
import torch
def _get_config_directory():
""" Find the predefined detector config directory """
try:
# Assume we are running in the source mmdetection repo
repo_dpath = dirname(dirname(__file__))
except NameError:
# For IPython development when this __file__ is not defined
import mmdet
repo_dpath = dirname(dirname(mmdet.__file__))
config_dpath = join(repo_dpath, 'configs')
if not exists(config_dpath):
raise Exception('Cannot find config path')
return config_dpath
def _get_config_module(fname):
"""
Load a configuration as a python module
"""
from mmcv import Config
config_dpath = _get_config_directory()
config_fpath = join(config_dpath, fname)
config_mod = Config.fromfile(config_fpath)
return config_mod
def _get_detector_cfg(fname):
"""
Grab configs necessary to create a detector. These are deep copied to allow
for safe modification of parameters without influencing other tests.
"""
import mmcv
config = _get_config_module(fname)
model = copy.deepcopy(config.model)
train_cfg = mmcv.Config(copy.deepcopy(config.train_cfg))
test_cfg = mmcv.Config(copy.deepcopy(config.test_cfg))
return model, train_cfg, test_cfg
def test_faster_rcnn_forward():
_test_two_stage_forward('nus/faster_rcnn_r50_fpn_caffe_2x8_1x_nus.py')
def _test_two_stage_forward(cfg_file):
model, train_cfg, test_cfg = _get_detector_cfg(cfg_file)
model['pretrained'] = None
from mmdet.models import build_detector
detector = build_detector(model, train_cfg=train_cfg, test_cfg=test_cfg)
input_shape = (1, 3, 256, 256)
# Test forward train with a non-empty truth batch
mm_inputs = _demo_mm_inputs(input_shape, num_items=[10])
imgs = mm_inputs.pop('imgs')
img_metas = mm_inputs.pop('img_metas')
gt_bboxes = mm_inputs['gt_bboxes']
gt_labels = mm_inputs['gt_labels']
gt_masks = mm_inputs['gt_masks']
losses = detector.forward(
imgs,
img_metas,
gt_bboxes=gt_bboxes,
gt_labels=gt_labels,
gt_masks=gt_masks,
return_loss=True)
assert isinstance(losses, dict)
from mmdet.apis.train import parse_losses
total_loss = parse_losses(losses)[0].requires_grad_(True)
assert float(total_loss.item()) > 0
total_loss.backward()
# Test forward train with an empty truth batch
mm_inputs = _demo_mm_inputs(input_shape, num_items=[0])
imgs = mm_inputs.pop('imgs')
img_metas = mm_inputs.pop('img_metas')
gt_bboxes = mm_inputs['gt_bboxes']
gt_labels = mm_inputs['gt_labels']
gt_masks = mm_inputs['gt_masks']
losses = detector.forward(
imgs,
img_metas,
gt_bboxes=gt_bboxes,
gt_labels=gt_labels,
gt_masks=gt_masks,
return_loss=True)
assert isinstance(losses, dict)
from mmdet.apis.train import parse_losses
total_loss = parse_losses(losses)[0].requires_grad_(True)
assert float(total_loss.item()) > 0
total_loss.backward()
# Test forward test
with torch.no_grad():
img_list = [g[None, :] for g in imgs]
batch_results = []
for one_img, one_meta in zip(img_list, img_metas):
result = detector.forward([one_img], [[one_meta]],
return_loss=False)
batch_results.append(result)
def _test_single_stage_forward(cfg_file):
model, train_cfg, test_cfg = _get_detector_cfg(cfg_file)
model['pretrained'] = None
from mmdet.models import build_detector
detector = build_detector(model, train_cfg=train_cfg, test_cfg=test_cfg)
input_shape = (1, 3, 300, 300)
mm_inputs = _demo_mm_inputs(input_shape)
imgs = mm_inputs.pop('imgs')
img_metas = mm_inputs.pop('img_metas')
# Test forward train
gt_bboxes = mm_inputs['gt_bboxes']
gt_labels = mm_inputs['gt_labels']
losses = detector.forward(
imgs,
img_metas,
gt_bboxes=gt_bboxes,
gt_labels=gt_labels,
return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
img_list = [g[None, :] for g in imgs]
batch_results = []
for one_img, one_meta in zip(img_list, img_metas):
result = detector.forward([one_img], [[one_meta]],
return_loss=False)
batch_results.append(result)
def _demo_mm_inputs(input_shape=(1, 3, 300, 300),
num_items=None, num_classes=10): # yapf: disable
"""
Create a superset of inputs needed to run test or train batches.
Args:
input_shape (tuple):
input batch dimensions
num_items (None | List[int]):
specifies the number of boxes in each batch item
num_classes (int):
number of different labels a box might have
"""
from mmdet.core import BitmapMasks
(N, C, H, W) = input_shape
rng = np.random.RandomState(0)
imgs = rng.rand(*input_shape)
img_metas = [{
'img_shape': (H, W, C),
'ori_shape': (H, W, C),
'pad_shape': (H, W, C),
'filename': '<demo>.png',
'scale_factor': 1.0,
'flip': False,
} for _ in range(N)]
gt_bboxes = []
gt_labels = []
gt_masks = []
for batch_idx in range(N):
if num_items is None:
num_boxes = rng.randint(1, 10)
else:
num_boxes = num_items[batch_idx]
cx, cy, bw, bh = rng.rand(num_boxes, 4).T
tl_x = ((cx * W) - (W * bw / 2)).clip(0, W)
tl_y = ((cy * H) - (H * bh / 2)).clip(0, H)
br_x = ((cx * W) + (W * bw / 2)).clip(0, W)
br_y = ((cy * H) + (H * bh / 2)).clip(0, H)
boxes = np.vstack([tl_x, tl_y, br_x, br_y]).T
class_idxs = rng.randint(1, num_classes, size=num_boxes)
gt_bboxes.append(torch.FloatTensor(boxes))
gt_labels.append(torch.LongTensor(class_idxs))
mask = np.random.randint(0, 2, (len(boxes), H, W), dtype=np.uint8)
gt_masks.append(BitmapMasks(mask, H, W))
mm_inputs = {
'imgs': torch.FloatTensor(imgs).requires_grad_(True),
'img_metas': img_metas,
'gt_bboxes': gt_bboxes,
'gt_labels': gt_labels,
'gt_bboxes_ignore': None,
'gt_masks': gt_masks,
}
return mm_inputs
tools/test.py
View file @ 6d71b439
...@@ -51,12 +51,12 @@ def parse_args(): ...@@ -51,12 +51,12 @@ def parse_args():
parser.add_argument('checkpoint', help='checkpoint file') parser.add_argument('checkpoint', help='checkpoint file')
parser.add_argument('--out', help='output result file in pickle format') parser.add_argument('--out', help='output result file in pickle format')
parser.add_argument( parser.add_argument(
'--fuse_conv_bn', '--fuse-conv-bn',
action='store_true', action='store_true',
help='Whether to fuse conv and bn, this will slightly increase' help='Whether to fuse conv and bn, this will slightly increase'
'the inference speed') 'the inference speed')
parser.add_argument( parser.add_argument(
'--format_only', '--format-only',
action='store_true', action='store_true',
help='Format the output results without perform evaluation. It is' help='Format the output results without perform evaluation. It is'
'useful when you want to format the result to a specific format and ' 'useful when you want to format the result to a specific format and '
...@@ -69,7 +69,7 @@ def parse_args(): ...@@ -69,7 +69,7 @@ def parse_args():
' "segm", "proposal" for COCO, and "mAP", "recall" for PASCAL VOC') ' "segm", "proposal" for COCO, and "mAP", "recall" for PASCAL VOC')
parser.add_argument('--show', action='store_true', help='show results') parser.add_argument('--show', action='store_true', help='show results')
parser.add_argument( parser.add_argument(
'--gpu_collect', '--gpu-collect',
action='store_true', action='store_true',
help='whether to use gpu to collect results.') help='whether to use gpu to collect results.')
parser.add_argument( parser.add_argument(
......
tools/train.py
View file @ 6d71b439
...@@ -11,19 +11,18 @@ from mmcv import Config ...@@ -11,19 +11,18 @@ from mmcv import Config
from mmcv.runner import init_dist from mmcv.runner import init_dist
from mmdet3d import __version__ from mmdet3d import __version__
from mmdet3d.apis import train_detector
from mmdet3d.datasets import build_dataset from mmdet3d.datasets import build_dataset
from mmdet3d.models import build_detector from mmdet3d.models import build_detector
from mmdet3d.utils import collect_env from mmdet3d.utils import collect_env
from mmdet.apis import get_root_logger, set_random_seed from mmdet.apis import get_root_logger, set_random_seed, train_detector
def parse_args(): def parse_args():
parser = argparse.ArgumentParser(description='Train a detector') parser = argparse.ArgumentParser(description='Train a detector')
parser.add_argument('config', help='train config file path') parser.add_argument('config', help='train config file path')
parser.add_argument('--work_dir', help='the dir to save logs and models') parser.add_argument('--work-dir', help='the dir to save logs and models')
parser.add_argument( parser.add_argument(
'--resume_from', help='the checkpoint file to resume from') '--resume-from', help='the checkpoint file to resume from')
parser.add_argument( parser.add_argument(
'--validate', '--validate',
action='store_true', action='store_true',
......
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