Skip to content

GitLab

Commit ecc6e480 authored by Benjamin Thomas Graham's avatar Benjamin Thomas Graham
Browse files

fixes

parent e488fe04
Hide whitespace changes
Inline Side-by-side
Showing with 200 additions and 264 deletions
examples/3d_segmentation/fully_convolutional.py
View file @ ecc6e480
...@@ -112,7 +112,7 @@ def iou(stats): ...@@ -112,7 +112,7 @@ def iou(stats):
# loop over CAD models # loop over CAD models
for k in range(len(pred)): for k in range(len(pred)):
p = pred[k] p = pred[k]
iou_per_part[k, j] = (inter(p, gt[k], j+1) + eps) / (union(p, gt[k], j+1) + eps) iou_per_part[k, j] = (inter(p, gt[k], j) + eps) / (union(p, gt[k], j) + eps)
# average over CAD models and parts # average over CAD models and parts
iou_all[i] = np.mean(iou_per_part) iou_all[i] = np.mean(iou_per_part)
# weighted average over categories # weighted average over categories
......
examples/3d_segmentation/unet.py
View file @ ecc6e480
...@@ -112,7 +112,7 @@ def iou(stats): ...@@ -112,7 +112,7 @@ def iou(stats):
# loop over CAD models # loop over CAD models
for k in range(len(pred)): for k in range(len(pred)):
p = pred[k] p = pred[k]
iou_per_part[k, j] = (inter(p, gt[k], j+1) + eps) / (union(p, gt[k], j+1) + eps) iou_per_part[k, j] = (inter(p, gt[k], j) + eps) / (union(p, gt[k], j) + eps)
# average over CAD models and parts # average over CAD models and parts
iou_all[i] = np.mean(iou_per_part) iou_all[i] = np.mean(iou_per_part)
# weighted average over categories # weighted average over categories
......
examples/nyu2/data/prepare_data.py deleted 100644 → 0
View file @ e488fe04
import numpy as np
import torch
import glob, math, os
import scipy.io
import h5py
import pickle
classes = [
'wall', 'floor', 'cabinet', 'bed',
'chair', 'sofa', 'table', 'door',
'window', 'bookshelf', 'picture', 'counter',
'blinds', 'desk', 'shelves', 'curtain',
'dresser', 'pillow', 'mirror', 'floor mat',
'clothes', 'ceiling', 'books', 'refridgerator',
'television', 'paper', 'towel', 'shower curtain',
'box', 'whiteboard', 'person', 'night stand',
'toilet', 'sink', 'lamp', 'bathtub',
'bag', 'otherstructure', 'otherfurniture', 'otherprop']
corresponding_classes_in_Silberman_labeling = [40, 40, 3, 22, 5, 40, 12, 38, 40, 40, 2, 39, 40, 40, 26, 40, 24,
40, 7, 40, 1, 40, 40, 34, 38, 29, 40, 8, 40, 40, 40, 40, 38, 40,
40, 14, 40, 38, 40, 40, 40, 15, 39, 40, 30, 40, 40, 39, 40, 39, 38,
40, 38, 40, 37, 40, 38, 38, 9, 40, 40, 38, 40, 11, 38, 40, 40, 40,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 38, 13, 40, 40, 6, 40, 23,
40, 39, 10, 16, 40, 40, 40, 40, 38, 40, 40, 40, 40, 40, 40, 40, 40,
40, 38, 40, 39, 40, 40, 40, 40, 39, 38, 40, 40, 40, 40, 40, 40, 18,
40, 40, 19, 28, 33, 40, 40, 40, 40, 40, 40, 40, 40, 40, 38, 27, 36,
40, 40, 40, 40, 21, 40, 20, 35, 40, 40, 40, 40, 40, 40, 40, 40, 38,
40, 40, 40, 4, 32, 40, 40, 39, 40, 39, 40, 40, 40, 40, 40, 17, 40,
40, 25, 40, 39, 40, 40, 40, 40, 40, 40, 40, 40, 39, 40, 40, 40, 40,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 39,
40, 40, 40, 40, 40, 40, 40, 40, 40, 39, 38, 38, 40, 40, 39, 40, 39,
40, 38, 39, 38, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 39, 40, 38,
40, 40, 38, 38, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
38, 40, 40, 40, 40, 40, 39, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
40, 40, 40, 40, 40, 39, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
40, 39, 40, 40, 40, 38, 40, 40, 39, 40, 40, 38, 40, 40, 40, 40, 40,
40, 40, 40, 40, 40, 40, 39, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
40, 40, 40, 40, 40, 40, 40, 31, 40, 40, 40, 40, 40, 40, 40, 38, 40,
40, 38, 39, 39, 40, 40, 40, 40, 40, 40, 40, 40, 40, 38, 40, 39, 40,
40, 39, 40, 40, 40, 38, 40, 40, 40, 40, 40, 40, 40, 40, 38, 39, 40,
40, 40, 40, 40, 40, 38, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
38, 39, 40, 40, 40, 40, 40, 40, 40, 39, 40, 40, 40, 40, 40, 40, 38,
40, 40, 40, 38, 40, 39, 40, 40, 40, 39, 39, 40, 40, 40, 40, 40, 40,
40, 40, 40, 40, 39, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
39, 39, 40, 40, 39, 39, 40, 40, 40, 40, 38, 40, 40, 38, 39, 39, 40,
39, 40, 39, 38, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 39, 40,
38, 40, 39, 40, 40, 40, 40, 40, 39, 39, 40, 40, 40, 40, 40, 40, 39,
39, 40, 40, 38, 39, 39, 40, 40, 40, 40, 40, 40, 40, 40, 40, 39, 39,
40, 40, 40, 40, 39, 40, 40, 40, 40, 40, 39, 40, 40, 39, 40, 40, 40,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 39, 38, 40, 40, 40,
40, 40, 40, 40, 39, 38, 39, 40, 38, 39, 40, 39, 40, 39, 40, 40, 40,
40, 40, 40, 40, 40, 38, 40, 40, 40, 40, 40, 38, 40, 40, 39, 40, 40,
40, 39, 40, 38, 40, 40, 40, 40, 40, 40, 40, 40, 38, 40, 40, 40, 40,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 39, 38, 40, 40, 38,
40, 40, 38, 40, 40, 40, 40, 40, 40, 40, 40, 40, 39, 40, 40, 40, 40,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 38, 40, 40, 38, 40,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 38, 38, 38, 40, 40, 40, 38,
40, 40, 40, 38, 38, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
40, 40, 40, 40, 40, 38, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
40, 40, 38, 40, 38, 39, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
40, 40, 39, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
40, 40, 40, 40, 39, 40, 39, 40, 40, 40, 40, 38, 38, 40, 40, 40, 38,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 39, 40, 40,
39, 40, 40, 39, 39, 40, 40, 40, 40, 40, 40, 40, 40, 39, 39, 39, 40,
40, 40, 40, 39, 40, 40, 40, 40, 40, 40, 40, 40, 39, 40, 40, 40, 40,
40, 39, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 38,
40, 40, 40, 40, 40, 40, 40, 39, 40, 40, 38, 40, 39, 40, 40, 40, 40,
38, 40, 40, 40, 40, 40, 38, 40, 40, 40, 40, 40, 40, 40, 39, 40, 40,
40, 40, 40, 40, 40, 40, 40, 39, 40, 40]
print(len(classes),len(corresponding_classes_in_Silberman_labeling))
split=scipy.io.loadmat('splits.mat')['testNdxs']-1 # 0-index
testIdxs=[x for x in range(1449) if x in split]
trainIdxs=[x for x in range(1449) if x not in split]
print(len(trainIdxs),len(testIdxs))
f = h5py.File('nyu_depth_v2_labeled.mat','r')
for i,x in enumerate(trainIdxs):
tc=f.get('images')[x]
td=f.get('depths')[x]*100
td-=td.mean()
print(td.std())
gt=np.array(f.get('labels')[x],dtype='int16')-1
coords=[]
col=[]
cl=[]
for x in range(40,600):
for y in range(45,470):
if gt[x,y]>=0:
cl.append(corresponding_classes_in_Silberman_labeling[gt[x,y]]-1)
coords.append([x-320,y-240,td[x,y]])
col.append([255,tc[0,x,y],tc[1,x,y],tc[2,x,y]])
coords=np.array(coords,dtype='int16')
col=np.array(col,dtype='uint8')
cl=np.array(cl,dtype='int8')
print(coords.shape,col.shape,cl.shape)
pickle.dump([coords,col,cl],open('train'+str(i)+'.pickle','wb'),protocol=pickle.HIGHEST_PROTOCOL)
f = h5py.File('nyu_depth_v2_labeled.mat','r')
for i,x in enumerate(testIdxs):
tc=f.get('images')[x]
td=f.get('depths')[x]*100
td-=td.mean()
print(td.std())
gt=np.array(f.get('labels')[x],dtype='int16')-1
coords=[]
col=[]
cl=[]
for x in range(40,600):
for y in range(45,470):
if gt[x,y]>=0:
cl.append(corresponding_classes_in_Silberman_labeling[gt[x,y]]-1)
coords.append([x-320,y-240,td[x,y]])
col.append([255,tc[0,x,y],tc[1,x,y],tc[2,x,y]])
coords=np.array(coords,dtype='int16')
col=np.array(col,dtype='uint8')
cl=np.array(cl,dtype='int8')
print(coords.shape,col.shape,cl.shape)
pickle.dump([coords,col,cl],open('test'+str(i)+'.pickle','wb'),protocol=pickle.HIGHEST_PROTOCOL)
sparseconvnet/SCN/CUDA/Convolution.cpp
View file @ ecc6e480
...@@ -32,11 +32,11 @@ double cuda_Convolution_updateOutput( ...@@ -32,11 +32,11 @@ double cuda_Convolution_updateOutput(
auto _rules = auto _rules =
m.getRuleBook(inputSize, outputSize, filterSize, filterStride, true); m.getRuleBook(inputSize, outputSize, filterSize, filterStride, true);
Int nActiveOut = m.getNActive(outputSize); Int nActiveOut = m.getNActive(outputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActiveOut, op});
if (nActiveOut) { if (nActiveOut) {
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActiveOut, op});
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto oF = output_features.data<T>(); auto oF = output_features.data<T>();
auto w = weight.data<T>(); auto w = weight.data<T>();
...@@ -67,12 +67,12 @@ void cuda_Convolution_backward( ...@@ -67,12 +67,12 @@ void cuda_Convolution_backward(
m.getRuleBook(inputSize, outputSize, filterSize, filterStride, true); m.getRuleBook(inputSize, outputSize, filterSize, filterStride, true);
Int nActiveIn = m.getNActive(inputSize); Int nActiveIn = m.getNActive(inputSize);
Int nActiveOut = m.getNActive(outputSize); Int nActiveOut = m.getNActive(outputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActiveIn, ip});
d_input_features.zero_();
if (nActiveOut) { if (nActiveOut) {
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActiveIn, ip});
d_input_features.zero_();
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto diF = d_input_features.data<T>(); auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>(); auto doF = d_output_features.data<T>();
...@@ -98,11 +98,11 @@ double cuda_SubmanifoldConvolution_updateOutput( ...@@ -98,11 +98,11 @@ double cuda_SubmanifoldConvolution_updateOutput(
auto _rules = m.getSubmanifoldRuleBook(inputSize, filterSize, true); auto _rules = m.getSubmanifoldRuleBook(inputSize, filterSize, true);
Int nActive = m.getNActive(inputSize); Int nActive = m.getNActive(inputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActive, op});
if (nActive) { if (nActive) {
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActive, op});
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto oF = output_features.data<T>(); auto oF = output_features.data<T>();
auto w = weight.data<T>(); auto w = weight.data<T>();
...@@ -130,12 +130,12 @@ void cuda_SubmanifoldConvolution_backward( ...@@ -130,12 +130,12 @@ void cuda_SubmanifoldConvolution_backward(
auto _rules = m.getSubmanifoldRuleBook(inputSize, filterSize, true); auto _rules = m.getSubmanifoldRuleBook(inputSize, filterSize, true);
Int nActive = m.getNActive(inputSize); Int nActive = m.getNActive(inputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActive, ip});
d_input_features.zero_();
if (nActive) { if (nActive) {
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActive, ip});
d_input_features.zero_();
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto diF = d_input_features.data<T>(); auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>(); auto doF = d_output_features.data<T>();
...@@ -160,11 +160,11 @@ double cuda_PermutohedralSubmanifoldConvolution_updateOutput( ...@@ -160,11 +160,11 @@ double cuda_PermutohedralSubmanifoldConvolution_updateOutput(
auto _rules = m.getPermutohedralSubmanifoldRuleBook(inputSize, true); auto _rules = m.getPermutohedralSubmanifoldRuleBook(inputSize, true);
Int nActive = m.getNActive(inputSize); Int nActive = m.getNActive(inputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActive, op});
if (nActive) { if (nActive) {
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActive, op});
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto oF = output_features.data<T>(); auto oF = output_features.data<T>();
auto w = weight.data<T>(); auto w = weight.data<T>();
...@@ -191,12 +191,12 @@ void cuda_PermutohedralSubmanifoldConvolution_backward( ...@@ -191,12 +191,12 @@ void cuda_PermutohedralSubmanifoldConvolution_backward(
auto _rules = m.getPermutohedralSubmanifoldRuleBook(inputSize, true); auto _rules = m.getPermutohedralSubmanifoldRuleBook(inputSize, true);
Int nActive = m.getNActive(inputSize); Int nActive = m.getNActive(inputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActive, ip});
d_input_features.zero_();
if (nActive) { if (nActive) {
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActive, ip});
d_input_features.zero_();
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto diF = d_input_features.data<T>(); auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>(); auto doF = d_output_features.data<T>();
...@@ -225,10 +225,11 @@ double cuda_FullConvolution_updateOutput( ...@@ -225,10 +225,11 @@ double cuda_FullConvolution_updateOutput(
auto _rules = mIn.getFullConvolutionRuleBook(inputSize, outputSize, auto _rules = mIn.getFullConvolutionRuleBook(inputSize, outputSize,
filterSize, filterStride, mOut); filterSize, filterStride, mOut);
Int nActiveOut = mOut.getNActive(outputSize); Int nActiveOut = mOut.getNActive(outputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActiveOut, op});
if (nActiveOut) { if (nActiveOut) {
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActiveOut, op});
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto oF = output_features.data<T>(); auto oF = output_features.data<T>();
auto w = weight.data<T>(); auto w = weight.data<T>();
...@@ -260,12 +261,12 @@ void cuda_FullConvolution_backward( ...@@ -260,12 +261,12 @@ void cuda_FullConvolution_backward(
filterSize, filterStride, mOut); filterSize, filterStride, mOut);
Int nActiveIn = mIn.getNActive(inputSize); Int nActiveIn = mIn.getNActive(inputSize);
Int nActiveOut = mOut.getNActive(outputSize); Int nActiveOut = mOut.getNActive(outputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActiveIn, ip});
d_input_features.zero_();
if (nActiveOut) { if (nActiveOut) {
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActiveIn, ip});
d_input_features.zero_();
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto diF = d_input_features.data<T>(); auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>(); auto doF = d_output_features.data<T>();
...@@ -292,11 +293,11 @@ double cuda_RandomizedStrideConvolution_updateOutput( ...@@ -292,11 +293,11 @@ double cuda_RandomizedStrideConvolution_updateOutput(
auto _rules = m.getRandomizedStrideRuleBook(inputSize, outputSize, filterSize, auto _rules = m.getRandomizedStrideRuleBook(inputSize, outputSize, filterSize,
filterStride, true); filterStride, true);
Int nActiveOut = m.getNActive(outputSize); Int nActiveOut = m.getNActive(outputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActiveOut, op});
if (nActiveOut) { if (nActiveOut) {
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActiveOut, op});
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto oF = output_features.data<T>(); auto oF = output_features.data<T>();
auto w = weight.data<T>(); auto w = weight.data<T>();
...@@ -327,12 +328,12 @@ void cuda_RandomizedStrideConvolution_backward( ...@@ -327,12 +328,12 @@ void cuda_RandomizedStrideConvolution_backward(
filterStride, true); filterStride, true);
Int nActiveIn = m.getNActive(inputSize); Int nActiveIn = m.getNActive(inputSize);
Int nActiveOut = m.getNActive(outputSize); Int nActiveOut = m.getNActive(outputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActiveIn, ip});
d_input_features.zero_();
if (nActiveOut) { if (nActiveOut) {
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActiveIn, ip});
d_input_features.zero_();
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto diF = d_input_features.data<T>(); auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>(); auto doF = d_output_features.data<T>();
......
sparseconvnet/SCN/CUDA/Deconvolution.cpp
View file @ ecc6e480
...@@ -28,11 +28,11 @@ double cuda_Deconvolution_updateOutput( ...@@ -28,11 +28,11 @@ double cuda_Deconvolution_updateOutput(
auto _rules = auto _rules =
m.getRuleBook(outputSize, inputSize, filterSize, filterStride, true); m.getRuleBook(outputSize, inputSize, filterSize, filterStride, true);
Int nActiveOut = m.getNActive(outputSize); Int nActiveOut = m.getNActive(outputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActiveOut, op});
if (nActiveOut) { if (nActiveOut) {
Int ip = weight.size(1);
Int op = weight.size(2);
output_features.resize_({nActiveOut, op});
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto oF = output_features.data<T>(); auto oF = output_features.data<T>();
auto w = weight.data<T>(); auto w = weight.data<T>();
...@@ -63,12 +63,12 @@ void cuda_Deconvolution_backward( ...@@ -63,12 +63,12 @@ void cuda_Deconvolution_backward(
m.getRuleBook(outputSize, inputSize, filterSize, filterStride, true); m.getRuleBook(outputSize, inputSize, filterSize, filterStride, true);
Int nActiveIn = m.getNActive(inputSize); Int nActiveIn = m.getNActive(inputSize);
Int nActiveOut = m.getNActive(outputSize); Int nActiveOut = m.getNActive(outputSize);
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActiveIn, ip});
d_input_features.zero_();
if (nActiveOut) { if (nActiveOut) {
Int ip = weight.size(1);
Int op = weight.size(2);
d_input_features.resize_({nActiveIn, ip});
d_input_features.zero_();
auto iF = input_features.data<T>(); auto iF = input_features.data<T>();
auto diF = d_input_features.data<T>(); auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>(); auto doF = d_output_features.data<T>();
......
sparseconvnet/SCN/Metadata/IOLayersRules.h
View file @ ecc6e480
...@@ -7,7 +7,6 @@ ...@@ -7,7 +7,6 @@
#ifndef INPUTLAYER_H #ifndef INPUTLAYER_H
#define INPUTLAYER_H #define INPUTLAYER_H
// Rulebook Format // Rulebook Format
// rules[0][0] == mode // rules[0][0] == mode
// rules[0][1] == maxActive per spatial location (==1 for modes 0,1,2) // rules[0][1] == maxActive per spatial location (==1 for modes 0,1,2)
...@@ -169,75 +168,119 @@ void blRules(SparseGrids<dimension> &SGs, RuleBook &rules, long *coords, ...@@ -169,75 +168,119 @@ void blRules(SparseGrids<dimension> &SGs, RuleBook &rules, long *coords,
return; return;
} }
// Compile list of how input rows correspond to output rows if (mode <= 2) {
std::vector<std::vector<std::vector<Int>>> outputRows(batchSize); // Compile list of how input rows correspond to output rows
std::vector<Int> nActives(batchSize); std::vector<std::vector<Int>> outputRows(batchSize);
std::vector<Int> nActives(batchSize);
#pragma omp parallel for private(I) #pragma omp parallel for private(I)
for (I = 0; I < batchSize; I++) { for (I = 0; I < batchSize; I++) {
auto &sg = SGs[I]; auto &sg = SGs[I];
auto &ors = outputRows[I]; auto &ors = outputRows[I];
auto &nAct = nActives[I]; auto &nAct = nActives[I];
auto c = coords + I * length * dimension; auto c = coords + I * length * dimension;
Int i = I * length; Int i = I * length;
Point<dimension> p; Point<dimension> p;
for (Int l = 0; l < length; ++l, ++i) {
for (Int j = 0; j < dimension; ++j) if (mode == 1) {
p[j] = *c++; for (Int l = 0; l < length; ++l, ++i) {
if (p[0] >= 0) { for (Int j = 0; j < dimension; ++j)
auto iter = sg.mp.find(p); p[j] = *c++;
if (iter == sg.mp.end()) { if (p[0] >= 0) {
sg.mp[p] = nAct++; auto iter = sg.mp.find(p);
ors.resize(nAct); if (iter == sg.mp.end()) {
sg.mp[p] = nAct++;
ors.push_back(i);
} else {
ors[sg.mp[p]] = i;
}
}
}
}
if (mode == 2) {
for (Int l = 0; l < length; ++l, ++i) {
for (Int j = 0; j < dimension; ++j)
p[j] = *c++;
if (p[0] >= 0) {
auto iter = sg.mp.find(p);
if (iter == sg.mp.end()) {
sg.mp[p] = nAct++;
ors.push_back(i);
}
}
} }
ors[sg.mp[p]].push_back(i);
} }
} }
}
for (I = 0; I < batchSize; I++) {
SGs[I].ctr = nActive;
nActive += nActives[I];
}
Int maxActive = 1;
if (mode >= 3)
for (auto &ors : outputRows)
for (auto &row : ors)
maxActive = std::max(maxActive, (Int)row.size());
rules.resize(2);
rules[0].push_back(mode);
rules[0].push_back(maxActive);
rules[0].push_back(batchSize);
rules[0].push_back(length);
rules[0].push_back(nActive);
auto &rule = rules[1];
if (mode == 1) {
rule.resize(2 * nActive);
#pragma omp parallel for private(I)
for (I = 0; I < batchSize; I++) { for (I = 0; I < batchSize; I++) {
auto &ors = outputRows[I]; SGs[I].ctr = nActive;
auto rr = &rule[SGs[I].ctr * 2]; nActive += nActives[I];
for (auto &row : ors) { }
rr[0] = row.size(); Int maxActive = 1;
rr[1] = row.back(); rules.resize(2);
rr += 2; rules[0].push_back(mode);
rules[0].push_back(maxActive);
rules[0].push_back(batchSize);
rules[0].push_back(length);
rules[0].push_back(nActive);
auto &rule = rules[1];
if (mode == 1) {
rule.resize(2 * nActive);
#pragma omp parallel for private(I)
for (I = 0; I < batchSize; I++) {
auto &ors = outputRows[I];
auto rr = &rule[SGs[I].ctr * 2];
for (auto &row : ors) {
rr[0] = 1;
rr[1] = row;
rr += 2;
}
} }
} }
return;
} }
if (mode == 2) {
rule.resize(2 * nActive); if (mode == 3 or mode == 4) {
// Compile list of how input rows correspond to output rows
std::vector<std::vector<std::vector<Int>>> outputRows(batchSize);
std::vector<Int> nActives(batchSize);
#pragma omp parallel for private(I) #pragma omp parallel for private(I)
for (I = 0; I < batchSize; I++) { for (I = 0; I < batchSize; I++) {
auto &sg = SGs[I];
auto &ors = outputRows[I]; auto &ors = outputRows[I];
auto rr = &rule[SGs[I].ctr * 2]; auto &nAct = nActives[I];
for (auto &row : ors) { auto c = coords + I * length * dimension;
rr[0] = row.size(); Int i = I * length;
rr[1] = row.front(); Point<dimension> p;
rr += 2; for (Int l = 0; l < length; ++l, ++i) {
for (Int j = 0; j < dimension; ++j)
p[j] = *c++;
if (p[0] >= 0) {
auto iter = sg.mp.find(p);
if (iter == sg.mp.end()) {
sg.mp[p] = nAct++;
ors.resize(nAct);
}
ors[sg.mp[p]].push_back(i);
}
} }
} }
}
if (mode == 3 or mode == 4) { for (I = 0; I < batchSize; I++) {
SGs[I].ctr = nActive;
nActive += nActives[I];
}
Int maxActive = 1;
if (mode >= 3)
for (auto &ors : outputRows)
for (auto &row : ors)
maxActive = std::max(maxActive, (Int)row.size());
rules.resize(2);
rules[0].push_back(mode);
rules[0].push_back(maxActive);
rules[0].push_back(batchSize);
rules[0].push_back(length);
rules[0].push_back(nActive);
auto &rule = rules[1];
rule.resize((maxActive + 1) * nActive); rule.resize((maxActive + 1) * nActive);
#pragma omp parallel for private(I) #pragma omp parallel for private(I)
for (I = 0; I < batchSize; I++) { for (I = 0; I < batchSize; I++) {
......
sparseconvnet/SCN/Metadata/Metadata.cpp
View file @ ecc6e480
...@@ -255,29 +255,28 @@ void Metadata<dimension>::appendMetadata(Metadata<dimension> &mAdd, ...@@ -255,29 +255,28 @@ void Metadata<dimension>::appendMetadata(Metadata<dimension> &mAdd,
template <Int dimension> template <Int dimension>
std::vector<at::Tensor> std::vector<at::Tensor>
Metadata<dimension>::sparsifyCompare(Metadata<dimension> &mReference, Metadata<dimension>::sparsifyCompare(Metadata<dimension> &mGT,
/*long*/ at::Tensor spatialSize) { /*long*/ at::Tensor spatialSize) {
auto p = LongTensorToPoint<dimension>(spatialSize); auto p = LongTensorToPoint<dimension>(spatialSize);
at::Tensor gt = torch::zeros({nActive[p]}, at::kByte); at::Tensor gt = torch::zeros({nActive[p]}, at::kByte);
at::Tensor ref_map = torch::empty({mReference.nActive[p]}, at::kLong); at::Tensor ref_map = torch::/*empty*/ zeros({mGT.nActive[p]}, at::kLong);
auto gtPtr = (signed char *)gt.data_ptr(); //<signed char>(); long *ref_map_ptr = ref_map.data<long>();
// auto gtPtr = gt.data<signed char>(); unsigned char *gt_ptr = gt.data<unsigned char>();
auto &sgsReference = mReference.grids[p]; auto &sgsGT = mGT.grids[p];
auto &sgsFull = grids[p]; auto &sgsFull = grids[p];
Int batchSize = sgsFull.size(); Int batchSize = sgsFull.size();
Int sample; Int sample;
#pragma omp parallel for private(sample) #pragma omp parallel for private(sample)
for (sample = 0; sample < (Int)batchSize; ++sample) { for (sample = 0; sample < (Int)batchSize; ++sample) {
auto &sgReference = sgsReference[sample]; auto &sgGT = sgsGT[sample];
auto &sgFull = sgsFull[sample]; auto &sgFull = sgsFull[sample];
for (auto const &iter : sgFull.mp) { for (auto const &iter : sgGT.mp) {
bool gt_ = sgReference.mp.find(iter.first) != sgReference.mp.end(); auto f = sgFull.mp.find(iter.first);
if (gt_) { if (f == sgFull.mp.end())
ref_map[sgReference.mp[iter.first] + sgReference.ctr] = std::cout << __FILE__ << ":" << __LINE__ << std::endl;
iter.second + sgFull.ctr; ref_map_ptr[iter.second + sgGT.ctr] = f->second + sgFull.ctr;
gtPtr[iter.second + sgFull.ctr] = +1; gt_ptr[f->second + sgFull.ctr] = +1;
}
} }
} }
return {gt, ref_map}; return {gt, ref_map};
...@@ -561,6 +560,19 @@ RuleBook &Metadata<dimension>::getRandomizedStrideRuleBook( ...@@ -561,6 +560,19 @@ RuleBook &Metadata<dimension>::getRandomizedStrideRuleBook(
return rb; return rb;
} }
at::Tensor vvl2t(std::vector<std::vector<long>> v) {
long s = 0;
for (auto &x : v)
s += x.size();
at::Tensor t = torch::empty({s}, at::CPU(at::kLong));
long *p = t.data<long>();
for (auto &x : v) {
std::memcpy(p, &x[0], x.size() * sizeof(long));
p += x.size();
}
return t;
}
template <Int dimension> template <Int dimension>
std::vector<at::Tensor> std::vector<at::Tensor>
Metadata<dimension>::compareSparseHelper(Metadata<dimension> &mR, Metadata<dimension>::compareSparseHelper(Metadata<dimension> &mR,
...@@ -568,33 +580,44 @@ Metadata<dimension>::compareSparseHelper(Metadata<dimension> &mR, ...@@ -568,33 +580,44 @@ Metadata<dimension>::compareSparseHelper(Metadata<dimension> &mR,
auto p = LongTensorToPoint<dimension>(spatialSize); auto p = LongTensorToPoint<dimension>(spatialSize);
auto &sgsL = grids[p]; auto &sgsL = grids[p];
auto &sgsR = mR.grids[p]; auto &sgsR = mR.grids[p];
std::vector<long> cL, cR, L, R; Int bs = sgsL.size(), sample;
for (Int sample = 0; sample < (Int)sgsL.size(); ++sample) { std::vector<std::vector<long>> cL(bs), cR(bs), L(bs), R(bs);
#pragma omp parallel for private(sample)
for (sample = 0; sample < bs; ++sample) {
auto &sgL = sgsL[sample]; auto &sgL = sgsL[sample];
auto &sgR = sgsR[sample]; auto &sgR = sgsR[sample];
auto &cLs = cL[sample];
auto &cRs = cR[sample];
auto &Ls = L[sample];
auto &Rs = R[sample];
for (auto const &iter : sgL.mp) { for (auto const &iter : sgL.mp) {
if (sgR.mp.find(iter.first) == sgR.mp.end()) { if (sgR.mp.find(iter.first) == sgR.mp.end()) {
L.push_back(sgL.mp[iter.first] + sgL.ctr); Ls.push_back(sgL.mp[iter.first] + sgL.ctr);
} else { } else {
cL.push_back(sgL.mp[iter.first] + sgL.ctr); cLs.push_back(sgL.mp[iter.first] + sgL.ctr);
cR.push_back(sgR.mp[iter.first] + sgR.ctr); cRs.push_back(sgR.mp[iter.first] + sgR.ctr);
} }
} }
for (auto const &iter : sgR.mp) { for (auto const &iter : sgR.mp) {
if (sgL.mp.find(iter.first) == sgL.mp.end()) { if (sgL.mp.find(iter.first) == sgL.mp.end()) {
R.push_back(sgR.mp[iter.first] + sgR.ctr); Rs.push_back(sgR.mp[iter.first] + sgR.ctr);
} }
} }
} }
at::Tensor cL_ = torch::empty({(long)cL.size()}, at::CPU(at::kLong)); return {vvl2t(cL), vvl2t(cR), vvl2t(L), vvl2t(R)};
std::memcpy(cL_.data<long>(), &cL[0], cL.size() * sizeof(long)); }
at::Tensor cR_ = torch::empty({(long)cR.size()}, at::CPU(at::kLong));
std::memcpy(cR_.data<long>(), &cR[0], cR.size() * sizeof(long)); at::Tensor vvl2t_(std::vector<std::vector<Int>> v) {
at::Tensor L_ = torch::empty({(long)L.size()}, at::CPU(at::kLong)); long s = 0;
std::memcpy(L_.data<long>(), &L[0], L.size() * sizeof(long)); for (auto &x : v)
at::Tensor R_ = torch::empty({(long)R.size()}, at::CPU(at::kLong)); s += x.size();
std::memcpy(R_.data<long>(), &R[0], R.size() * sizeof(long)); at::Tensor t = torch::empty({s}, at::CPU(at_kINT));
return {cL_, cR_, L_, R_}; Int *p = t.data<Int>();
for (auto &x : v) {
std::memcpy(p, &x[0], x.size() * sizeof(Int));
p += x.size();
}
return t;
} }
template <Int dimension> Int volume(long *point) { template <Int dimension> Int volume(long *point) {
......
sparseconvnet/networkArchitectures.py
View file @ ecc6e480
# Copyright 2g016-present, Facebook, Inc. # Copyright 2016-present, Facebook, Inc.
# All rights reserved. # All rights reserved.
# #
# This source code is licensed under the BSD-style license found in the # This source code is licensed under the BSD-style license found in the
......
sparseconvnet/sparseConvNetTensor.py
View file @ ecc6e480
...@@ -36,10 +36,9 @@ class SparseConvNetTensor(object): ...@@ -36,10 +36,9 @@ class SparseConvNetTensor(object):
self.features = self.features.cpu() self.features = self.features.cpu()
return self return self
def set_(self): @property
self.features.set_(self.features.storage_type()()) def requires_grad(self):
self.metadata.set_() return self.features.requires_grad
self.spatialSize = None
def __repr__(self): def __repr__(self):
sl = self.get_spatial_locations() if self.metadata else None sl = self.get_spatial_locations() if self.metadata else None
...@@ -50,11 +49,3 @@ class SparseConvNetTensor(object): ...@@ -50,11 +49,3 @@ class SparseConvNetTensor(object):
',batch_locations.shape=' + repr(sl.shape if self.metadata else None) + \ ',batch_locations.shape=' + repr(sl.shape if self.metadata else None) + \
',spatial size=' + repr(self.spatial_size) + \ ',spatial size=' + repr(self.spatial_size) + \
'>>' '>>'
def to_variable(self, requires_grad=False, volatile=False):
"Convert self.features to a variable for use with modern PyTorch interface."
self.features = Variable(
self.features,
requires_grad=requires_grad,
volatile=volatile)
return self
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