-
Notifications
You must be signed in to change notification settings - Fork 4
/
train_test.py
412 lines (320 loc) · 19.1 KB
/
train_test.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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
import os
import time
import numpy as np
from scipy.spatial import cKDTree
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"]="0"
import tensorflow as tf
import scipy.io as sio
from model import *
#!/usr/bin/env python3
# -*- coding: utf-8 -*
flags = tf.app.flags
FLAGS = flags.FLAGS
# Training parameterss
flags.DEFINE_float('learning_rate', 1e-4, 'initial learning rate.')
flags.DEFINE_integer('batch_size', 8, 'batch size.')
# Architecture parameters
flags.DEFINE_integer('num_fclayers', 2, 'network depth')
flags.DEFINE_integer('num_evecs', 30, "number of eigenvectors used for representation")
flags.DEFINE_string('targets_dir', '../../Unsupervised_FMnet/Shapes/Surreal/MAT/','directory with shapes')
flags.DEFINE_string('feat_dir', '../../Unsupervised_FMnet/Shapes/Surreal/MAT/','directory with shapes')
flags.DEFINE_string('files_name', 'surreal_', 'name common to all the shapes')
flags.DEFINE_string('te_dir_F', '../../Unsupervised_FMnet/Shapes/Faust_r_aligned/MAT/','directory with shapes')
flags.DEFINE_string('feat_dir_te_F', '../../Unsupervised_FMnet/Shapes/Faust_r_aligned/faust_mat/','directory with shapes')
flags.DEFINE_string('files_name_F', '', 'name common to all the shapes')
flags.DEFINE_string('te_dir_S', '../../Unsupervised_FMnet/Shapes/Scape_r_aligned/MAT/','directory with shapes')
flags.DEFINE_string('feat_dir_te_S', '../../Unsupervised_FMnet/Shapes/Scape_r_aligned/scape_mat/','directory with shapes')
flags.DEFINE_string('files_name_S', '', 'name common to all the shapes')
flags.DEFINE_integer('max_train_iter', 10000, '')
flags.DEFINE_integer('num_vertices', 3000, '')
flags.DEFINE_integer('save_summaries_secs', 500, '')
flags.DEFINE_integer('save_model_secs', 500, '')
flags.DEFINE_string('log_dir_', 'Training/SCAPE_r_aligned/pointnet_fmnet_b8_lr_-4_30evec_.001_6k_2fc_aligned_tr-Sur_100_rand_sup_both_sides_1_1.001_sbb-/',
'directory to save models and results')
flags.DEFINE_string('matches_dir_F', './Matches/FAUST_r_aligned/ptnet_surreal_80rand_3k_F_30_1_1_.001_b8_both_te_20/','directory to save models and results')
flags.DEFINE_string('matches_dir_S', './Matches/SCAPE_r_aligned/ptnet_surreal_80_rand_3k_S_30_1_1_.001_b8_both_te_20/','directory to save models and results')
flags.DEFINE_integer('dim_', 128,'')
flags.DEFINE_integer('decay_step', 200000, help='Decay step for lr decay [default: 200000]')
flags.DEFINE_float('decay_rate', 0.7, help='Decay rate for lr decay [default: 0.7]')
# Globals
dim_=FLAGS.dim_
flags.DEFINE_list('dims', [dim_,dim_,dim_,dim_, dim_, dim_, dim_], '')
dim_1_layer = int(FLAGS.dims[0])
flags.DEFINE_integer('dim_shot', dim_1_layer, '')
no_layers = FLAGS.num_fclayers
last_layer = int(FLAGS.dims[no_layers-1])
flags.DEFINE_integer('dim_out', last_layer, '')
vert_dir = FLAGS.feat_dir
vert_dir_te_S = FLAGS.feat_dir_te_S
vert_dir_te_F = FLAGS.feat_dir_te_F
n_tr = 100
#train_subjects = list(range(n_tr))
train_subjects = np.random.choice(1000,n_tr)
test_subjects_F,test_subjects_S = (range(80,100),range(52,70))
main_dir = FLAGS.targets_dir
files_name =FLAGS.files_name
te_files_name_F = FLAGS.files_name_F
te_files_name_S = FLAGS.files_name_S
#te_files_name = ''
test_dir_F=FLAGS.te_dir_F
test_dir_S=FLAGS.te_dir_S
DECAY_STEP = FLAGS.decay_step
DECAY_RATE = FLAGS.decay_rate
BN_INIT_DECAY = 0.5
BN_DECAY_DECAY_RATE = 0.5
BN_DECAY_DECAY_STEP = float(DECAY_STEP)
BN_DECAY_CLIP = 0.99
BATCH_SIZE = FLAGS.batch_size
def get_input_pair(batch_size, num_vertices, dataset):
batch_input = {
'source_evecs': np.zeros((batch_size, num_vertices, FLAGS.num_evecs)),
'target_evecs': np.zeros((batch_size, num_vertices, FLAGS.num_evecs)),
'source_evecs_trans': np.zeros((batch_size,FLAGS.num_evecs,num_vertices)),
'target_evecs_trans': np.zeros((batch_size,FLAGS.num_evecs,num_vertices)),
'source_shot': np.zeros((batch_size, num_vertices, 3)),
'target_shot': np.zeros((batch_size, num_vertices, 3)),
'source_evals': np.zeros((batch_size, FLAGS.num_evecs)),
'target_evals': np.zeros((batch_size, FLAGS.num_evecs))
}
for i_batch in range(batch_size):
i_source = train_subjects[np.random.choice(range(n_tr))]
i_target = train_subjects[np.random.choice(range(n_tr))]
batch_input_ = get_pair_from_ram(i_target, i_source, dataset)
batch_input_['source_labels'] = range(np.shape(batch_input_['source_evecs'])[0])
batch_input_['target_labels'] = range(np.shape(batch_input_['target_evecs'])[0])
joint_lbls = np.intersect1d(batch_input_['source_labels'],batch_input_['target_labels'])
#print(joint_lbls)
joint_labels_source = np.random.permutation(joint_lbls)[:num_vertices]
joint_labels_target = np.random.permutation(joint_lbls)[:num_vertices]
ind_dict_source = {value: ind for ind, value in enumerate(batch_input_['source_labels'])}
ind_source = [ind_dict_source[x] for x in joint_labels_source]
ind_dict_target = {value: ind for ind, value in enumerate(batch_input_['target_labels'])}
ind_target = [ind_dict_target[x] for x in joint_labels_target]
message = "number of indices must be equal"
assert len(ind_source) == len(ind_target), message
evecs_full = batch_input_['source_evecs']
#print(evecs_full.shape)
evecs= evecs_full[ind_source, :]
evecs_trans = batch_input_['source_evecs_trans'][:, ind_source]
shot = batch_input_['source_shot'][ind_source, :]
#print(batch_input_['target_shot'].shape)
evals = [item for sublist in batch_input_['source_evals'] for item in sublist] # what?
batch_input['source_evecs'][i_batch] = evecs
batch_input['source_evecs_trans'][i_batch] = evecs_trans
batch_input['source_shot'][i_batch] = shot
batch_input['source_evals'][i_batch] = evals
evecs = batch_input_['target_evecs'][ind_target, :]
evecs_trans = batch_input_['target_evecs_trans'][:, ind_target]
shot = batch_input_['target_shot'][ind_target, :]
evals = [item for sublist in batch_input_['target_evals'] for item in sublist]
batch_input['target_evecs'][i_batch] = evecs
batch_input['target_evecs_trans'][i_batch] = evecs_trans
batch_input['target_shot'][i_batch] = shot
batch_input['target_evals'][i_batch] = evals
return batch_input
def get_input_pair_test(i_target, i_source, sub_):
batch_input = {}
batch_input_ = get_pair_from_ram(i_target, i_source, sub_)
evecs = batch_input_['source_evecs']
evecs_trans = batch_input_['source_evecs_trans']
shot = batch_input_['source_shot']
evals = [item for sublist in batch_input_['source_evals'] for item in sublist]
batch_input['source_evecs'] = evecs
batch_input['source_evecs_trans'] = evecs_trans
batch_input['source_shot'] = shot
batch_input['source_evals'] = evals
evecs = batch_input_['target_evecs']
evecs_trans = batch_input_['target_evecs_trans']
shot = batch_input_['target_shot']
evals = [item for sublist in batch_input_['target_evals'] for item in sublist]
batch_input['target_evecs'] = evecs
batch_input['target_evecs_trans'] = evecs_trans
batch_input['target_shot'] = shot
batch_input['target_evals'] = evals
return batch_input
def get_pair_from_ram(i_target, i_source, sub_):
input_data = {}
if sub_ == 'train':
targets_= targets_train
elif sub_== 'te_F':
targets_=targets_test_F
else:
targets_=targets_test_S
evecs = targets_[i_source]['target_evecs']
evecs_trans = targets_[i_source]['target_evecs_trans']
shot = targets_[i_source]['target_shot']
evals = targets_[i_source]['target_evals']
input_data['source_evecs'] = evecs
input_data['source_evecs_trans'] = evecs_trans
input_data['source_shot'] = shot
input_data['source_evals'] = evals
input_data.update(targets_[i_target])
return input_data
def load_targets_to_ram():
global targets_train,targets_test_F,targets_test_S
targets_train,targets_test_F,targets_test_S = ({},{},{})
targets_train = load_subs(train_subjects, main_dir, vert_dir,files_name)
targets_test_F= load_subs(test_subjects_F,test_dir_F, vert_dir_te_F, te_files_name_F)
targets_test_S =load_subs(test_subjects_S,test_dir_S, vert_dir_te_S,te_files_name_S)
def load_subs(subjects_list, dir_name,v_dir,f_name):
targets = {}
for i_target in subjects_list:
target_file = dir_name +f_name +'%.4d.mat' % (i_target)
vert_file = v_dir +f_name +'%.4d.mat' % (i_target)
#print(vert_file)
input_data = sio.loadmat(target_file)
evecs = input_data['target_evecs'][:, 0:FLAGS.num_evecs]
evecs_trans = input_data['target_evecs_trans'][0:FLAGS.num_evecs,:]
evals = input_data['target_evals'][0:FLAGS.num_evecs]
input_data['target_evecs'] = evecs
input_data['target_evecs_trans'] = evecs_trans
input_data['target_evals'] = evals
p_feat = sio.loadmat(vert_file)
input_data['target_shot'] =[]
input_data['target_shot'] = p_feat['VERT']
targets[i_target] = input_data
return targets
def get_bn_decay(batch):
bn_momentum = tf.train.exponential_decay(
BN_INIT_DECAY,
batch*BATCH_SIZE,
BN_DECAY_DECAY_STEP,
BN_DECAY_DECAY_RATE,
staircase=True)
bn_decay = tf.minimum(BN_DECAY_CLIP, 1 - bn_momentum)
return bn_decay
def run_training():
print('log_dir=%s' % FLAGS.log_dir_)
if not os.path.isdir(FLAGS.log_dir_):
os.makedirs(FLAGS.log_dir_)
print('building graph...')
with tf.Graph().as_default():
# Set placeholders for inputs
source_shot = tf.placeholder(tf.float32,shape=(None, None, 3),name='source_shot')
target_shot = tf.placeholder(tf.float32, shape=(None, None, 3),name='target_shot')
source_evecs = tf.placeholder(tf.float32, shape=(None, None, FLAGS.num_evecs), name='source_evecs')
source_evecs_trans = tf.placeholder(tf.float32,shape=(None, FLAGS.num_evecs, None),name='source_evecs_trans')
source_evals = tf.placeholder(tf.float32,shape=(None, FLAGS.num_evecs),name='source_evals')
target_evecs = tf.placeholder(tf.float32,shape=(None, None, FLAGS.num_evecs),name='target_evecs')
target_evecs_trans = tf.placeholder(tf.float32,shape=(None, FLAGS.num_evecs, None),name='target_evecs_trans')
target_evals = tf.placeholder(tf.float32,shape=(None, FLAGS.num_evecs),name='target_evals')
# train\test switch flag
phase = tf.placeholder(dtype=tf.bool, name='phase')
#is_training_pl = tf.placeholder(tf.bool, shape=())
#print (is_training_pl)
batch = tf.Variable(0)
bn_decay = get_bn_decay(batch)
net_loss, safeguard_inverse, net,end_points, C, merged = get_model(phase, source_shot, target_shot, source_evecs, source_evecs_trans,
source_evals, target_evecs, target_evecs_trans, target_evals, bn_decay)
summary = tf.summary.scalar("num_evecs", float(FLAGS.num_evecs))
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
global_step = tf.Variable(0, name='global_step', trainable=False)
optimizer = tf.train.AdamOptimizer(FLAGS.learning_rate)
train_op = optimizer.minimize(net_loss,global_step=global_step, aggregation_method=2)
#saver = tf.train.Saver(max_to_keep=40)
saver = tf.train.Saver(tf.global_variables())
sv = tf.train.Supervisor(logdir=FLAGS.log_dir_,init_op=tf.global_variables_initializer(),local_init_op=tf.local_variables_initializer(),
global_step=global_step,save_summaries_secs=FLAGS.save_summaries_secs,
save_model_secs=FLAGS.save_model_secs,summary_op=None,saver=saver)
writer = sv.summary_writer
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
print('starting session...')
iteration = 0
with sv.managed_session(config=config) as sess:
print('loading data to ram...')
load_targets_to_ram()
print('starting training loop...')
while not sv.should_stop() and iteration < FLAGS.max_train_iter:
iteration += 1
start_time = time.time()
input_data = get_input_pair(FLAGS.batch_size, FLAGS.num_vertices, 'train')
feed_dict = {phase: False, source_shot: input_data['source_shot'], target_shot: input_data['target_shot'],
source_evecs: input_data['source_evecs'], source_evecs_trans: input_data['source_evecs_trans'],
source_evals: input_data['source_evals'], target_evecs: input_data['target_evecs'],
target_evecs_trans: input_data['target_evecs_trans'], target_evals: input_data['target_evals']}
summaries, step, my_loss, safeguard, _ = sess.run([merged, global_step, net_loss, safeguard_inverse, train_op],
feed_dict=feed_dict)
writer.add_summary(summaries, step)
summary_ = sess.run(summary)
writer.add_summary(summary_, step)
duration = time.time() - start_time
print('train - step %d: loss = %.2f (%.3f sec)'% (step, my_loss, duration))
if iteration%1000==0:
for i_source in range(52,69):
for i_target in range(i_source+1,70):
t = time.time()
input_data = get_input_pair_test(i_target, i_source, 'te_S')
source_evecs_ = input_data['source_evecs'][:, 0:FLAGS.num_evecs]
target_evecs_ = input_data['target_evecs'][:, 0:FLAGS.num_evecs]
feed_dict = {
phase: False,
source_shot: [input_data['source_shot']],
target_shot: [input_data['target_shot']],
source_evecs: [input_data['source_evecs']],
source_evecs_trans: [input_data['source_evecs_trans']],
source_evals: [input_data['source_evals']],
target_evecs: [input_data['target_evecs']],
target_evecs_trans: [input_data['target_evecs_trans']],
target_evals: [input_data['target_evals']]
}
C_est_ = sess.run([C], feed_dict=feed_dict)
Ct = np.squeeze(C_est_).T #Keep transposed
kdt = cKDTree(np.matmul(source_evecs_, Ct))
dist, indices = kdt.query(target_evecs_, n_jobs=-1)
indices = indices + 1
print("Computed correspondences for pair: %s, %s." % (i_source, i_target) +
" Took %f seconds" % (time.time() - t))
params_to_save = {}
params_to_save['matches'] = indices
#params_to_save['C'] = Ct.T
# For Matlab where index start at 1
new_dir = FLAGS.matches_dir_S + '%.3d-' % iteration + '/'
if not os.path.isdir(new_dir):
print('matches_dir=%s' % new_dir)
os.makedirs(new_dir)
sio.savemat(new_dir + '%.4d-' % i_source + '%.4d.mat' % i_target, params_to_save)
for i_source in range(80,99):
for i_target in range(i_source+1,100):
t = time.time()
input_data = get_input_pair_test(i_target, i_source, 'te_F')
source_evecs_ = input_data['source_evecs'][:, 0:FLAGS.num_evecs]
target_evecs_ = input_data['target_evecs'][:, 0:FLAGS.num_evecs]
feed_dict = {
phase: False,
source_shot: [input_data['source_shot']],
target_shot: [input_data['target_shot']],
source_evecs: [input_data['source_evecs']],
source_evecs_trans: [input_data['source_evecs_trans']],
source_evals: [input_data['source_evals']],
target_evecs: [input_data['target_evecs']],
target_evecs_trans: [input_data['target_evecs_trans']],
target_evals: [input_data['target_evals']]
}
C_est_ = sess.run([C], feed_dict=feed_dict)
Ct = np.squeeze(C_est_).T #Keep transposed
kdt = cKDTree(np.matmul(source_evecs_, Ct))
dist, indices = kdt.query(target_evecs_, n_jobs=-1)
indices = indices + 1
print("Computed correspondences for pair: %s, %s." % (i_source, i_target) +
" Took %f seconds" % (time.time() - t))
params_to_save = {}
params_to_save['matches'] = indices
#params_to_save['C'] = Ct.T
# For Matlab where index start at 1
new_dir = FLAGS.matches_dir_F + '%.3d-' % iteration + '/'
if not os.path.isdir(new_dir):
print('matches_dir=%s' % new_dir)
os.makedirs(new_dir)
sio.savemat(new_dir + '%.4d-' % i_source + '%.4d.mat' % i_target, params_to_save)
def main(_):
import time
start_time = time.time()
run_training()
print("--- %s seconds ---" % (time.time() - start_time))
if __name__ == '__main__':
tf.app.run()