-
Notifications
You must be signed in to change notification settings - Fork 1
/
evaluation.py
458 lines (402 loc) · 20.1 KB
/
evaluation.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
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
"""Evaluation"""
from __future__ import print_function
import os
import sys
import torch
import numpy as np
from model.UGNCL import UGNCL
from data import get_test_loader
from vocab import deserialize_vocab
from collections import OrderedDict
class AverageMeter(object):
"""Computes and stores the average and current value"""
def __init__(self):
self.reset()
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=0):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / (.0001 + self.count)
def __str__(self):
"""String representation for logging
"""
# for values that should be recorded exactly e.g. iteration number
if self.count == 0:
return str(self.val)
# for stats
return '%.4f (%.4f)' % (self.val, self.avg)
class LogCollector(object):
"""A collection of logging objects that can change from train to val"""
def __init__(self):
# to keep the order of logged variables deterministic
self.meters = OrderedDict()
def update(self, k, v, n=0):
# create a new meter if previously not recorded
if k not in self.meters:
self.meters[k] = AverageMeter()
self.meters[k].update(v, n)
def __str__(self):
"""Concatenate the meters in one log line
"""
s = ''
for i, (k, v) in enumerate(self.meters.items()):
if i > 0:
s += ' '
s += k + ' ' + str(v)
return s
def tb_log(self, tb_logger, prefix='', step=None):
"""Log using tensorboard
"""
for k, v in self.meters.items():
tb_logger.log_value(prefix + k, v.val, step=step)
def encode_data(model, data_loader, log_step=10, logging=print, sub=0):
"""Encode all images and captions loadable by `data_loader`
"""
val_logger = LogCollector()
# switch to evaluate mode
model.val_start()
# np array to keep all the embeddings
img_embs1 = None
img_embs2 = None
cap_embs = None
max_n_word = 0
for i, (images, captions, lengths, ids, _) in enumerate(data_loader):
max_n_word = max(max_n_word, max(lengths))
ids_ = []
for i, (images, captions, lengths, ids, _) in enumerate(data_loader):
# make sure val logger is used
model.logger = val_logger
ids_ += ids
# compute the embeddings
with torch.no_grad():
img_emb1, img_emb2, cap_emb, cap_len = model.forward_emb(images, captions, lengths)
if cap_embs is None:
img_embs1 = np.zeros((len(data_loader.dataset), img_emb1.size(1), img_emb1.size(2)))
img_embs2 = np.zeros((len(data_loader.dataset), img_emb2.size(1), img_emb2.size(2)))
cap_embs = np.zeros((len(data_loader.dataset), max_n_word, cap_emb.size(2)))
cap_lens = [0] * len(data_loader.dataset)
# cache embeddings
img_embs1[ids, :, :] = img_emb1.data.cpu().numpy().copy()
img_embs2[ids, :, :] = img_emb2.data.cpu().numpy().copy()
cap_embs[ids, :max(lengths), :] = cap_emb.data.cpu().numpy().copy()
for j, nid in enumerate(ids):
cap_lens[nid] = cap_len[j]
del images, captions
if sub > 0:
print(f"===>batch {i}")
if sub > 0 and i > sub:
break
if sub > 0:
return np.array(img_embs1)[ids_].tolist(), np.array(cap_embs)[ids_].tolist(), np.array(cap_lens)[
ids_].tolist(), ids_
else:
return img_embs1, img_embs2, cap_embs, cap_lens
def shard_attn_scores(model, img_embs, cap_embs, cap_lens, opt, shard_size=100, mode="sim"):
n_im_shard = (len(img_embs) - 1) // shard_size + 1
n_cap_shard = (len(cap_embs) - 1) // shard_size + 1
sims = np.zeros((len(img_embs), len(cap_embs)))
for i in range(n_im_shard):
im_start, im_end = shard_size * i, min(shard_size * (i + 1), len(img_embs))
for j in range(n_cap_shard):
sys.stdout.write('\r>> shard_attn_scores batch (%d,%d)' % (i, j))
ca_start, ca_end = shard_size * j, min(shard_size * (j + 1), len(cap_embs))
with torch.no_grad():
im = torch.from_numpy(img_embs[im_start:im_end]).float().cuda()
ca = torch.from_numpy(cap_embs[ca_start:ca_end]).float().cuda()
l = cap_lens[ca_start:ca_end]
if mode == "sim":
sim = model.forward_sim(im, ca, l, mode)
else:
_, sim, _ = model.forward_sim(im, ca, l, mode) # Calculate evidence for retrieval
sims[im_start:im_end, ca_start:ca_end] = sim.data.cpu().numpy()
sys.stdout.write('\n')
return sims
def t2i(npts, sims, per_captions=1, return_ranks=False):
"""
Text->Images (Image Search)
Images: (N, n_region, d) matrix of images
Captions: (per_captions * N, max_n_word, d) matrix of captions
CapLens: (per_captions * N) array of caption lengths
sims: (N, per_captions * N) matrix of similarity im-cap
"""
ranks = np.zeros(per_captions * npts)
top1 = np.zeros(per_captions * npts)
top5 = np.zeros((per_captions * npts, 5), dtype=int)
# --> (per_captions * N(caption), N(image))
sims = sims.T
retreivaled_index = []
for index in range(npts):
for i in range(per_captions):
inds = np.argsort(sims[per_captions * index + i])[::-1]
retreivaled_index.append(inds)
ranks[per_captions * index + i] = np.where(inds == index)[0][0]
top1[per_captions * index + i] = inds[0]
top5[per_captions * index + i] = inds[0:5]
# Compute metrics
r1 = 100.0 * len(np.where(ranks < 1)[0]) / len(ranks)
r5 = 100.0 * len(np.where(ranks < 5)[0]) / len(ranks)
r10 = 100.0 * len(np.where(ranks < 10)[0]) / len(ranks)
medr = np.floor(np.median(ranks)) + 1
meanr = ranks.mean() + 1
if return_ranks:
return (r1, r5, r10, medr, meanr), (ranks, top1, top5, retreivaled_index)
else:
return (r1, r5, r10, medr, meanr)
def i2t(npts, sims, per_captions=1, return_ranks=False):
"""
Images->Text (Image Annotation)
Images: (N, n_region, d) matrix of images
Captions: (per_captions * N, max_n_word, d) matrix of captions
CapLens: (per_captions * N) array of caption lengths
sims: (N, per_captions * N) matrix of similarity im-cap
"""
ranks = np.zeros(npts)
top1 = np.zeros(npts)
top5 = np.zeros((npts, 5), dtype=int)
retreivaled_index = []
for index in range(npts):
inds = np.argsort(sims[index])[::-1]
retreivaled_index.append(inds)
# Score
rank = 1e20
for i in range(per_captions * index, per_captions * index + per_captions, 1):
tmp = np.where(inds == i)[0][0]
if tmp < rank:
rank = tmp
ranks[index] = rank
top1[index] = inds[0]
top5[index] = inds[0:5]
# Compute metrics
r1 = 100.0 * len(np.where(ranks < 1)[0]) / len(ranks)
r5 = 100.0 * len(np.where(ranks < 5)[0]) / len(ranks)
r10 = 100.0 * len(np.where(ranks < 10)[0]) / len(ranks)
medr = np.floor(np.median(ranks)) + 1
meanr = ranks.mean() + 1
if return_ranks:
return (r1, r5, r10, medr, meanr), (ranks, top1, top5, retreivaled_index)
else:
return (r1, r5, r10, medr, meanr)
def validation(opt, val_loader, model, fold=False):
# compute the encoding for all the validation images and captions
if opt.data_name == 'cc152k_precomp':
per_captions = 1
elif opt.data_name in ['coco_precomp', 'f30k_precomp']:
per_captions = 5
else:
print(f"No dataset")
return 0
model.val_start()
print('Encoding with model')
img_embs1, img_embs2, cap_embs, cap_lens = encode_data(model.similarity_model, val_loader)
# clear duplicate 5*images and keep 1*images FIXME
if not fold:
img_embs1 = np.array([img_embs1[i] for i in range(0, len(img_embs1), per_captions)])
img_embs2 = np.array([img_embs2[i] for i in range(0, len(img_embs2), per_captions)])
# record computation time of validation
print('Computing similarity from model')
sims_mean1 = shard_attn_scores(model.similarity_model, img_embs1, cap_embs, cap_lens, opt, shard_size=1000,
mode="not sim")
sims_mean2 = shard_attn_scores(model.similarity_model, img_embs2, cap_embs, cap_lens, opt, shard_size=1000,
mode="not sim")
sims_mean = (sims_mean1 + sims_mean2) / 2
print("Calculate similarity time with model")
(r1, r5, r10, medr, meanr) = i2t(img_embs1.shape[0], sims_mean, per_captions, return_ranks=False)
print("Average i2t Recall: %.2f" % ((r1 + r5 + r10) / 3))
print("Image to text: %.2f, %.2f, %.2f, %.2f, %.2f" % (r1, r5, r10, medr, meanr))
# image retrieval
(r1i, r5i, r10i, medri, meanr) = t2i(img_embs1.shape[0], sims_mean, per_captions, return_ranks=False)
print("Average t2i Recall: %.2f" % ((r1i + r5i + r10i) / 3))
print("Text to image: %.2f, %.2f, %.2f, %.2f, %.2f" % (r1i, r5i, r10i, medri, meanr))
r_sum = r1 + r5 + r10 + r1i + r5i + r10i
print("Sum of Recall: %.2f" % (r_sum))
else:
# 5fold cross-validation, only for MSCOCO
results = []
for i in range(5):
img_embs_shard1 = img_embs1[i * 5000:(i + 1) * 5000:5]
img_embs_shard2 = img_embs2[i * 5000:(i + 1) * 5000:5]
cap_embs_shard = cap_embs[i * 5000:(i + 1) * 5000]
cap_lens_shard = cap_lens[i * 5000:(i + 1) * 5000]
sims1 = shard_attn_scores(model.similarity_model, img_embs_shard1, cap_embs_shard, cap_lens_shard, opt,
shard_size=1000,
mode="not sim")
sims2 = shard_attn_scores(model.similarity_model, img_embs_shard2, cap_embs_shard, cap_lens_shard, opt,
shard_size=1000,
mode="not sim")
sims = (sims1 + sims2) / 2
print('Computing similarity from model')
r, rt = i2t(img_embs_shard1.shape[0], sims, per_captions, return_ranks=True)
ri, rti = t2i(img_embs_shard1.shape[0], sims, per_captions, return_ranks=True)
print("Image to text: %.1f, %.1f, %.1f, %.1f, %.1f" % r)
print("Text to image: %.1f, %.1f, %.1f, %.1f, %.1f" % ri)
ar = (r[0] + r[1] + r[2]) / 3
ari = (ri[0] + ri[1] + ri[2]) / 3
rsum = r[0] + r[1] + r[2] + ri[0] + ri[1] + ri[2]
print("rsum: %.1f ar: %.1f ari: %.1f" % (rsum, ar, ari))
results += [list(r) + list(ri) + [ar, ari, rsum]]
print("-----------------------------------")
print("Mean metrics: ")
mean_metrics = tuple(np.array(results).mean(axis=0).flatten())
a = np.array(mean_metrics)
print("Average i2t Recall: %.1f" % mean_metrics[11])
print("Image to text: %.1f %.1f %.1f %.1f %.1f" %
mean_metrics[:5])
print("Average t2i Recall: %.1f" % mean_metrics[12])
print("Text to image: %.1f %.1f %.1f %.1f %.1f" %
mean_metrics[5:10])
print("rsum: %.1f" % (a[0:3].sum() + a[5:8].sum()))
def validation_dul(opt, val_loader, models, fold=False):
# compute the encoding for all the validation images and captions
if opt.data_name == 'cc152k_precomp':
per_captions = 1
elif opt.data_name in ['coco_precomp', 'f30k_precomp']:
per_captions = 5
else:
print(f"No dataset")
return 0
models[0].val_start()
models[1].val_start()
print('Encoding with model')
img_embs0_1, img_embs0_2, cap_embs0, cap_lens0 = encode_data(models[0].similarity_model, val_loader, opt.log_step)
img_embs1_1, img_embs1_2, cap_embs1, cap_lens1 = encode_data(models[1].similarity_model, val_loader, opt.log_step)
if not fold:
img_embs0_1 = np.array([img_embs0_1[i] for i in range(0, len(img_embs0_1), per_captions)])
img_embs0_2 = np.array([img_embs0_2[i] for i in range(0, len(img_embs0_2), per_captions)])
img_embs1_1 = np.array([img_embs1_1[i] for i in range(0, len(img_embs1_1), per_captions)])
img_embs1_2 = np.array([img_embs1_2[i] for i in range(0, len(img_embs1_2), per_captions)])
# record computation time of validation
print('Computing similarity from model')
sims_mean = shard_attn_scores(models[0].similarity_model, img_embs0_1, cap_embs0, cap_lens0, opt, shard_size=1000,
mode="not sim")
sims_mean += shard_attn_scores(models[0].similarity_model, img_embs0_2, cap_embs0, cap_lens0, opt,
shard_size=1000, mode="not sim")
sims_mean += shard_attn_scores(models[1].similarity_model, img_embs1_1, cap_embs1, cap_lens1, opt,
shard_size=1000, mode="not sim")
sims_mean += shard_attn_scores(models[1].similarity_model, img_embs1_2, cap_embs1, cap_lens1, opt,
shard_size=1000, mode="not sim")
sims_mean /= 4
print("Calculate similarity time with model")
# caption retrieval
(r1, r5, r10, medr, meanr) = i2t(img_embs0_1.shape[0], sims_mean, per_captions, return_ranks=False)
print("Average i2t Recall: %.2f" % ((r1 + r5 + r10) / 3))
print("Image to text: %.2f, %.2f, %.2f, %.2f, %.2f" % (r1, r5, r10, medr, meanr))
# image retrieval
(r1i, r5i, r10i, medri, meanr) = t2i(img_embs0_1.shape[0], sims_mean, per_captions, return_ranks=False)
print("Average t2i Recall: %.2f" % ((r1i + r5i + r10i) / 3))
print("Text to image: %.2f, %.2f, %.2f, %.2f, %.2f" % (r1i, r5i, r10i, medri, meanr))
r_sum = r1 + r5 + r10 + r1i + r5i + r10i
print("Sum of Recall: %.2f" % (r_sum))
return r_sum
else:
# 5fold cross-validation, only for MS-COCO
results = []
for i in range(5):
img_embs_shard0_1 = img_embs0_1[i * 5000:(i + 1) * 5000:5]
img_embs_shard0_2 = img_embs0_2[i * 5000:(i + 1) * 5000:5]
cap_embs_shard = cap_embs0[i * 5000:(i + 1) * 5000]
cap_lens_shard = cap_lens0[i * 5000:(i + 1) * 5000]
img_embs_shard1_1 = img_embs1_1[i * 5000:(i + 1) * 5000:5]
img_embs_shard1_2 = img_embs1_2[i * 5000:(i + 1) * 5000:5]
cap_embs_shard1 = cap_embs1[i * 5000:(i + 1) * 5000]
cap_lens_shard1 = cap_lens1[i * 5000:(i + 1) * 5000]
sims = shard_attn_scores(models[0].similarity_model, img_embs_shard0_1, cap_embs_shard, cap_lens_shard, opt,
shard_size=1000,
mode="not sim")
sims += shard_attn_scores(models[0].similarity_model, img_embs_shard0_2, cap_embs_shard, cap_lens_shard, opt,
shard_size=1000,
mode="not sim")
sims += shard_attn_scores(models[1].similarity_model, img_embs_shard1_1, cap_embs_shard1, cap_lens_shard1, opt,
shard_size=1000,
mode="not sim")
sims += shard_attn_scores(models[1].similarity_model, img_embs_shard1_2, cap_embs_shard1, cap_lens_shard1,
opt,
shard_size=1000,
mode="not sim")
sims /= 4
print('Computing similarity from model')
r, rt0 = i2t(img_embs_shard1_1.shape[0], sims, per_captions, return_ranks=True)
ri, rti0 = t2i(img_embs_shard1_1.shape[0], sims, per_captions, return_ranks=True)
print("Image to text: %.1f, %.1f, %.1f, %.1f, %.1f" % r)
print("Text to image: %.1f, %.1f, %.1f, %.1f, %.1f" % ri)
if i == 0:
rt, rti = rt0, rti0
ar = (r[0] + r[1] + r[2]) / 3
ari = (ri[0] + ri[1] + ri[2]) / 3
rsum = r[0] + r[1] + r[2] + ri[0] + ri[1] + ri[2]
print("rsum: %.1f ar: %.1f ari: %.1f" % (rsum, ar, ari))
results += [list(r) + list(ri) + [ar, ari, rsum]]
print("-----------------------------------")
print("Mean metrics: ")
mean_metrics = tuple(np.array(results).mean(axis=0).flatten())
a = np.array(mean_metrics)
print("Average i2t Recall: %.1f" % mean_metrics[11])
print("Image to text: %.1f %.1f %.1f %.1f %.1f" %
mean_metrics[:5])
print("Average t2i Recall: %.1f" % mean_metrics[12])
print("Text to image: %.1f %.1f %.1f %.1f %.1f" %
mean_metrics[5:10])
print("rsum: %.1f" % (a[0:3].sum() + a[5:8].sum()))
def eval_UGNCL(checkpoint_paths, avg_SGRAF=True, data_path=None, vocab_path=None):
if avg_SGRAF is False:
print(f"Load checkpoint from '{checkpoint_paths[0]}'")
checkpoint = torch.load(checkpoint_paths[0])
opt = checkpoint['opt']
print(
f"Noise ratio is {opt.noise_ratio}, module is {opt.module_name}, best validation epoch is {checkpoint['epoch']} ({checkpoint['best_rsum']})")
if vocab_path != None:
opt.vocab_path = vocab_path
if data_path != None:
opt.data_path = data_path
vocab = deserialize_vocab(os.path.join(opt.vocab_path, '%s_vocab.json' % opt.data_name))
model = UGNCL(opt)
model.load_state_dict(checkpoint['model'])
if 'coco' in opt.data_name:
test_loader = get_test_loader('testall', opt.data_name, vocab, 100, 0, opt)
validation(opt, test_loader, model=model, fold=True)
validation(opt, test_loader, model=model, fold=False)
else:
test_loader = get_test_loader('test', opt.data_name, vocab, 100, 0, opt)
validation(opt, test_loader, model=model, fold=False)
else:
assert len(checkpoint_paths) == 2
print(f"Load checkpoint from '{checkpoint_paths}'")
checkpoint0 = torch.load(checkpoint_paths[0])
checkpoint1 = torch.load(checkpoint_paths[1])
opt0 = checkpoint0['opt']
opt1 = checkpoint1['opt']
print(
f"Noise ratios are {opt0.noise_ratio} and {opt1.noise_ratio}, "
f"modules are {opt0.module_name} and {opt1.module_name}, best validation epochs are {checkpoint0['epoch']}"
f" ({checkpoint0['best_rsum']}) and {checkpoint1['epoch']} ({checkpoint1['best_rsum']})")
vocab = deserialize_vocab(os.path.join(vocab_path, '%s_vocab.json' % opt0.data_name))
model0 = UGNCL(opt0)
model1 = UGNCL(opt1)
model0.load_state_dict(checkpoint0['model'])
model1.load_state_dict(checkpoint1['model'])
if 'coco' in opt0.data_name:
testall_loader = get_test_loader('testall', opt0.data_name, vocab, 100, 0, opt0)
print(f'=====>model {opt0.module_name} fold:True')
validation(opt0, testall_loader, model0, fold=False)
print(f'=====>model {opt1.module_name} fold:True')
validation(opt0, testall_loader, model1, fold=False)
print(f'=====>model SGRAF fold:True')
validation_dul(opt0, testall_loader, models=[model0, model1], fold=False)
print(f'=====>model {opt0.module_name} fold:False')
validation(opt0, test_loader, model0, fold=False)
print(f'=====>model {opt1.module_name} fold:False')
validation(opt0, test_loader, model1, fold=False)
print('=====>model SGRAF fold:False')
validation_dul(opt0, test_loader, models=[model0, model1], fold=False)
else:
test_loader = get_test_loader('test', opt0.data_name, vocab, 100, 0, opt0)
print(f'=====>model {opt0.module_name} fold:False')
validation(opt0, test_loader, model0, fold=False)
print(f'=====>model {opt1.module_name} fold:False')
validation(opt0, test_loader, model1, fold=False)
print('=====>model SGRAF fold:False')
validation_dul(opt0, test_loader, models=[model0, model1], fold=False)