-
Notifications
You must be signed in to change notification settings - Fork 1
/
framework.py
417 lines (325 loc) · 16.7 KB
/
framework.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
import os
import gc
import argparse
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.optim.lr_scheduler import MultiStepLR
from data_finetune import ModelNet40
from data_stanford import Stanford
from util import transform_point_cloud, npmat2euler, unsupervisedloss, supervisedloss, Chamfer_dis, Chamfer_distance, ab_angle
import numpy as np
from torch.utils.data import DataLoader
from tensorboardX import SummaryWriter
from tqdm import tqdm
from matching import matchingPerformance
from vis import visualize_pc
import time
import h5py
def test_one_epoch(args, net, test_loader):
net.eval()
total_loss = 0
num_examples = 0
rotations = []
translations = []
rotations_pred = []
translations_pred = []
corres = []
targets = []
transformed_src = []
Time_total = 0
Num = 0
i = 0
I_gts = []
transformed_src_gt = []
for src, target, rotation, translation, I_gt, _, _ in tqdm(test_loader):
src = src.cuda()
target = target.cuda()
rotation = rotation.cuda()
translation = translation.cuda()
batch_size = src.size(0)
num_examples += batch_size
torch.cuda.synchronize()
time_start=time.time()
rotation_pred, translation_pred, corre_src, scores = net(src, target)
torch.cuda.synchronize()
time_end=time.time()
diff_time = time_end - time_start
if i>10 and i<150:
Num = Num + batch_size
Time_total = Time_total+diff_time
i = i+1
## save rotation and translation
rotations.append(rotation.detach().cpu().numpy())
translations.append(translation.detach().cpu().numpy())
rotations_pred.append(rotation_pred.detach().cpu().numpy())
translations_pred.append(translation_pred.detach().cpu().numpy())
namta = 100.0
unloss = unsupervisedloss(src, corre_src)
suloss = supervisedloss(I_gt, scores)
###########################
loss = unloss + namta * suloss
#loss = namta * suloss
total_loss += loss.item() * batch_size
## for visualization
if args.eval:
transformed_src_batch = torch.matmul(rotation_pred, src) + translation_pred.unsqueeze(2)
transformed_src_batch_gt = torch.matmul(rotation, src) + translation.unsqueeze(2)
visualize_pc(src, target, target, transformed_src_batch, corre_src)
transformed_src.append(transformed_src_batch.detach().cpu().numpy())
transformed_src_gt.append(transformed_src_batch_gt.detach().cpu().numpy())
targets.append(target.detach().cpu().numpy())
corres.append(corre_src.detach().cpu().numpy())
I_gts.append(I_gt.detach().cpu().numpy())
#print("total time:{}; Num:{}; average: {}".format(Time_total, Num, Time_total/Num))
## for inliers ratio changes
if args.eval:
transformed_src = np.concatenate(transformed_src, axis=0)
transformed_src_gt = np.concatenate(transformed_src_gt, axis=0)
targets = np.concatenate(targets, axis=0)
corres = np.concatenate(corres, axis=0)
I_gts = np.concatenate(I_gts, axis=0)
num_ = int(I_gts.shape[0] / 10)
I_gts_ = I_gts[:(num_*10)]
inliers_ratio = np.sum(I_gts_)/(num_*10.0 *I_gts.shape[1])
print("ground truth inliers ratio: {}".format(inliers_ratio))
CD = Chamfer_distance(transformed_src.transpose((0,2,1)), corres.transpose((0,2,1)))
print("chamfer distance is: {}".format(CD))
CD2 = Chamfer_distance(transformed_src_gt.transpose((0,2,1)), targets.transpose((0,2,1)))
print("chamfer gt distance is: {}".format(CD2))
#matchingPerformance(transformed_src_gt, corres)
print("the following is real inliers ratio:")
#matchingPerformance(transformed_src_gt, targets)
print("the following is predict inliers ratio:")
#matchingPerformance(transformed_src, corres)
#matchingPerformance(transformed_src, targets)
rotations = np.concatenate(rotations, axis=0)
translations = np.concatenate(translations, axis=0)
rotations_pred = np.concatenate(rotations_pred, axis=0)
translations_pred = np.concatenate(translations_pred, axis=0)
return total_loss * 1.0 / num_examples, rotations, translations, rotations_pred, translations_pred
def test_one_epoch_for_sun3d(args, net, test_loader):
net.eval()
total_loss = 0
num_examples = 0
rotations = []
translations = []
rotations_pred = []
translations_pred = []
corres = []
targets = []
transformed_src = []
Time_total = 0
Num = 0
i = 0
I_gts = []
for src, target, rotation, translation, I_gt, idx1, idx2 in tqdm(test_loader):
src = src.cuda()
target = target.cuda()
rotation = rotation.cuda()
translation = translation.cuda()
batch_size = src.size(0)
num_examples += batch_size
torch.cuda.synchronize()
time_start=time.time()
rotation_pred, translation_pred, corre_src, scores = net(src, target)
torch.cuda.synchronize()
time_end=time.time()
diff_time = time_end - time_start
if i>10 and i<150:
Num = Num + batch_size
Time_total = Time_total+diff_time
i = i+1
if args.eval:
for i in range(batch_size):
name = str(idx1[i].detach().cpu().numpy())+'_'+str(idx2[i].detach().cpu().numpy())
save_path = os.path.join('./sun3d_pose_har/',name+'.h5')
with h5py.File(save_path, 'w') as h5file:
h5file.create_dataset('R', data=rotation_pred[i].detach().cpu().numpy(), compression="gzip", compression_opts=9)
h5file.create_dataset('t', data=translation_pred[i].detach().cpu().numpy(), compression="gzip", compression_opts=9)
print("save successfully")
## save rotation and translation
rotations.append(rotation.detach().cpu().numpy())
translations.append(translation.detach().cpu().numpy())
rotations_pred.append(rotation_pred.detach().cpu().numpy())
translations_pred.append(translation_pred.detach().cpu().numpy())
namta = 100.0
unloss = unsupervisedloss(src, corre_src)
suloss = supervisedloss(I_gt, scores)
###########################
loss = unloss + namta * suloss
#loss = namta * suloss
total_loss += loss.item() * batch_size
if args.eval:
transformed_src_batch = torch.matmul(rotation_pred, src) + translation_pred.unsqueeze(2)
#visualize_pc(src, target, target, transformed_src_batch, corre_src)
transformed_src.append(transformed_src_batch.detach().cpu().numpy())
targets.append(target.detach().cpu().numpy())
corres.append(corre_src.detach().cpu().numpy())
I_gts.append(I_gt.detach().cpu().numpy())
#print("total time:{}; Num:{}; average: {}".format(Time_total, Num, Time_total/Num))
if args.eval:
transformed_src = np.concatenate(transformed_src, axis=0)
targets = np.concatenate(targets, axis=0)
corres = np.concatenate(corres, axis=0)
I_gts = np.concatenate(I_gts, axis=0)
num_ = int(I_gts.shape[0] / 10)
I_gts_ = I_gts[:(num_*10)]
inliers_ratio = np.sum(I_gts_)/(num_*10.0 *I_gts.shape[1])
print("ground truth inliers ratio: {}".format(inliers_ratio))
CD = Chamfer_distance(transformed_src.transpose((0,2,1)), targets.transpose((0,2,1)))
print("chamfer distance is: {}".format(CD))
matchingPerformance(transformed_src, corres)
rotations = np.concatenate(rotations, axis=0)
translations = np.concatenate(translations, axis=0)
rotations_pred = np.concatenate(rotations_pred, axis=0)
translations_pred = np.concatenate(translations_pred, axis=0)
return total_loss * 1.0 / num_examples, rotations, translations, rotations_pred, translations_pred
def train_one_epoch(args, net, train_loader, opt):
net.train()
total_loss = 0
num_examples = 0
rotations = []
translations = []
rotations_pred = []
translations_pred = []
for src, target, rotation, translation, I_gt in tqdm(train_loader):
src = src.cuda()
target = target.cuda()
rotation = rotation.cuda()
translation = translation.cuda()
batch_size = src.size(0)
opt.zero_grad()
num_examples += batch_size
rotation_pred, translation_pred, corre_src, scores = net(src, target)
## save rotation and translation
rotations.append(rotation.detach().cpu().numpy())
translations.append(translation.detach().cpu().numpy())
rotations_pred.append(rotation_pred.detach().cpu().numpy())
translations_pred.append(translation_pred.detach().cpu().numpy())
# transformed_src = transform_point_cloud(src, rotation_ab_pred, translation_ab_pred)
# transformed_target = transform_point_cloud(target, rotation_ba_pred, translation_ba_pred)
namta = 100.0
unloss = unsupervisedloss(src, corre_src)
suloss = supervisedloss(I_gt, scores)
###########################
loss = unloss + namta * suloss
#loss = namta * suloss
loss.backward()
opt.step()
total_loss += loss.item() * batch_size
rotations = np.concatenate(rotations, axis=0)
translations = np.concatenate(translations, axis=0)
rotations_pred = np.concatenate(rotations_pred, axis=0)
translations_pred = np.concatenate(translations_pred, axis=0)
return total_loss * 1.0 / num_examples, rotations, translations, rotations_pred, translations_pred
def test(args, net, test_loader, boardio, textio):
with torch.no_grad():
test_loss, test_rotations, test_translations, test_rotations_pred, test_translations_pred = test_one_epoch(args, net, test_loader)
test_rotations_pred_euler = npmat2euler(test_rotations_pred)
test_rotations_euler = npmat2euler(test_rotations)
test_r_mse = np.mean((test_rotations_pred_euler - test_rotations_euler) ** 2)
test_r_rmse = np.sqrt(test_r_mse)
test_r_mae = np.mean(np.abs(test_rotations_pred_euler - test_rotations_euler))
test_t_mse = np.mean((test_translations - test_translations_pred) ** 2)
test_t_rmse = np.sqrt(test_t_mse)
test_t_mae = np.mean(np.abs(test_translations - test_translations_pred))
angle_error = ab_angle(test_rotations_pred, test_rotations)
textio.cprint('==FINAL TEST==')
textio.cprint('A--------->B')
textio.cprint('EPOCH:: %d, Loss: %f, rot_MSE: %f, rot_RMSE: %f, '
'rot_MAE: %f, trans_MSE: %f, trans_RMSE: %f, trans_MAE: %f, Angle_error: %f'
% (-1, test_loss, test_r_mse, test_r_rmse, test_r_mae, test_t_mse, test_t_rmse, test_t_mae, angle_error))
def train(args, net, train_loader, test_loader, boardio, textio):
if args.use_sgd:
print("Use SGD")
opt = optim.SGD(net.parameters(), lr=args.lr * 100, momentum=args.momentum, weight_decay=1e-4)
else:
print("Use Adam")
opt = optim.Adam(net.parameters(), lr=args.lr, weight_decay=1e-4)
scheduler = MultiStepLR(opt, milestones=[75, 150, 200], gamma=0.1)
best_test_loss = np.inf
best_test_r_mse = np.inf
best_test_r_rmse = np.inf
best_test_r_mae = np.inf
best_test_t_mse = np.inf
best_test_t_rmse = np.inf
best_test_t_mae = np.inf
for epoch in range(args.epochs):
scheduler.step()
train_loss, train_rotations, train_translations, train_rotations_pred, train_translations_pred = train_one_epoch(args, net, train_loader, opt)
with torch.no_grad():
test_loss, test_rotations, test_translations, test_rotations_pred, test_translations_pred = test_one_epoch(args, net, test_loader)
train_rotations_pred_euler = npmat2euler(train_rotations_pred)
train_rotations_euler = npmat2euler(train_rotations)
train_r_mse = np.mean((train_rotations_pred_euler - train_rotations_euler) ** 2)
train_r_rmse = np.sqrt(train_r_mse)
train_r_mae = np.mean(np.abs(train_rotations_pred_euler - train_rotations_euler))
train_t_mse = np.mean((train_translations - train_translations_pred) ** 2)
train_t_rmse = np.sqrt(train_t_mse)
train_t_mae = np.mean(np.abs(train_translations - train_translations_pred))
test_rotations_pred_euler = npmat2euler(test_rotations_pred)
test_rotations_euler = npmat2euler(test_rotations)
test_r_mse = np.mean((test_rotations_pred_euler - test_rotations_euler) ** 2)
test_r_rmse = np.sqrt(test_r_mse)
test_r_mae = np.mean(np.abs(test_rotations_pred_euler - test_rotations_euler))
test_t_mse = np.mean((test_translations - test_translations_pred) ** 2)
test_t_rmse = np.sqrt(test_t_mse)
test_t_mae = np.mean(np.abs(test_translations - test_translations_pred))
if best_test_loss >= test_loss:
best_test_loss = test_loss
best_test_r_mse = test_r_mse
best_test_r_rmse = test_r_rmse
best_test_r_mae = test_r_mae
best_test_t_mse = test_t_mse
best_test_t_rmse = test_t_rmse
best_test_t_mae = test_t_mae
if torch.cuda.device_count() > 1:
torch.save(net.module.state_dict(), 'checkpoints/%s/models/model.best.t7' % args.exp_name)
else:
torch.save(net.state_dict(), 'checkpoints/%s/models/model.best.t7' % args.exp_name)
textio.cprint('==TRAIN==')
textio.cprint('A--------->B')
textio.cprint('EPOCH:: %d, Loss: %f, rot_MSE: %f, rot_RMSE: %f, '
'rot_MAE: %f, trans_MSE: %f, trans_RMSE: %f, trans_MAE: %f'
% (epoch, train_loss, train_r_mse, train_r_rmse, train_r_mae, train_t_mse, train_t_rmse, train_t_mae))
textio.cprint('==TEST==')
textio.cprint('A--------->B')
textio.cprint('EPOCH:: %d, Loss: %f, rot_MSE: %f, rot_RMSE: %f, '
'rot_MAE: %f, trans_MSE: %f, trans_RMSE: %f, trans_MAE: %f'
% (epoch, test_loss, test_r_mse, test_r_rmse, test_r_mae, test_t_mse, test_t_rmse, test_t_mae))
textio.cprint('==BEST TEST==')
textio.cprint('A--------->B')
textio.cprint('EPOCH:: %d, Loss: %f, rot_MSE: %f, rot_RMSE: %f, '
'rot_MAE: %f, trans_MSE: %f, trans_RMSE: %f, trans_MAE: %f'
% (epoch, best_test_loss, best_test_r_mse, best_test_r_rmse,
best_test_r_mae, best_test_t_mse, best_test_t_rmse, best_test_t_mae))
boardio.add_scalar('A->B/train/loss', train_loss, epoch)
boardio.add_scalar('A->B/train/rotation/MSE', train_r_mse, epoch)
boardio.add_scalar('A->B/train/rotation/RMSE', train_r_rmse, epoch)
boardio.add_scalar('A->B/train/rotation/MAE', train_r_mae, epoch)
boardio.add_scalar('A->B/train/translation/MSE', train_t_mse, epoch)
boardio.add_scalar('A->B/train/translation/RMSE', train_t_rmse, epoch)
boardio.add_scalar('A->B/train/translation/MAE', train_t_mae, epoch)
############TEST
boardio.add_scalar('A->B/test/loss', test_loss, epoch)
boardio.add_scalar('A->B/test/rotation/MSE', test_r_mse, epoch)
boardio.add_scalar('A->B/test/rotation/RMSE', test_r_rmse, epoch)
boardio.add_scalar('A->B/test/rotation/MAE', test_r_mae, epoch)
boardio.add_scalar('A->B/test/translation/MSE', test_t_mse, epoch)
boardio.add_scalar('A->B/test/translation/RMSE', test_t_rmse, epoch)
boardio.add_scalar('A->B/test/translation/MAE', test_t_mae, epoch)
############BEST TEST
boardio.add_scalar('A->B/best_test/loss', best_test_loss, epoch)
boardio.add_scalar('A->B/best_test/rotation/MSE', best_test_r_mse, epoch)
boardio.add_scalar('A->B/best_test/rotation/RMSE', best_test_r_rmse, epoch)
boardio.add_scalar('A->B/best_test/rotation/MAE', best_test_r_mae, epoch)
boardio.add_scalar('A->B/best_test/translation/MSE', best_test_t_mse, epoch)
boardio.add_scalar('A->B/best_test/translation/RMSE', best_test_t_rmse, epoch)
boardio.add_scalar('A->B/best_test/translation/MAE', best_test_t_mae, epoch)
if torch.cuda.device_count() > 1:
torch.save(net.module.state_dict(), 'checkpoints/%s/models/model.%d.t7' % (args.exp_name, epoch))
else:
torch.save(net.state_dict(), 'checkpoints/%s/models/model.%d.t7' % (args.exp_name, epoch))
gc.collect()