-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
167 lines (137 loc) · 6.46 KB
/
utils.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
import os
import numpy as np
import h5py
import json
import torch
from scipy.misc import imread, imresize
from tqdm import tqdm
from collections import Counter
from random import seed, choice, sample
def create_input_files(dataset, karpathy_json_path, image_folder, captions_per_image, min_word_freq, output_folder,
max_len=100):
assert dataset in {'coco', 'flickr8k', 'flickr30k' }
with open(karpathy_json_path, 'r') as j:
data = json.load(j)
train_image_paths = []
train_image_captions = []
val_image_paths = []
val_image_captions = []
test_image_paths = []
test_image_captions = []
word_freq = Counter()
for img in data['images']:
captions = []
for c in img['sentences']:
word_freq.update(c['tokens'])
if len(c['tokens']) <= max_len:
captions.append(c['tokens'])
if len(captions) == 0:
continue
path = os.path.join(image_folder, img['filepath'], img['filename']) if dataset in ['coco'] else os.path.join(
image_folder, img['filename'])
if img['split'] in {'train', 'restval'}:
train_image_paths.append(path)
train_image_captions.append(captions)
elif img['split'] in {'val'}:
val_image_paths.append(path)
val_image_captions.append(captions)
elif img['split'] in {'test'}:
test_image_paths.append(path)
test_image_captions.append(captions)
assert len(train_image_paths) == len(train_image_captions)
assert len(val_image_paths) == len(val_image_captions)
assert len(test_image_paths) == len(test_image_captions)
words = [w for w in word_freq.keys() if word_freq[w] > min_word_freq]
word_map = {k: v + 1 for v, k in enumerate(words)}
word_map['<unk>'] = len(word_map) + 1
word_map['<start>'] = len(word_map) + 1
word_map['<end>'] = len(word_map) + 1
word_map['<pad>'] = 0
base_filename = dataset + '_' + str(captions_per_image) + '_cap_per_img_' + str(min_word_freq) + '_min_word_freq'
with open(os.path.join(output_folder, 'WORDMAP_' + base_filename + '.json'), 'w') as j:
json.dump(word_map, j)
seed(123)
for impaths, imcaps, split in [(train_image_paths, train_image_captions, 'TRAIN'),
(val_image_paths, val_image_captions, 'VAL'),
(test_image_paths, test_image_captions, 'TEST')]:
with h5py.File(os.path.join(output_folder, split + '_IMAGES_' + base_filename + '.hdf5'), 'a') as h:
h.attrs['captions_per_image'] = captions_per_image
images = h.create_dataset('images', (len(impaths), 3, 255, 255), dtype='uint8')
print("\nReading %s images and captions, storing to file...\n" % split)
enc_captions = []
caplens = []
image_paths = list()
for i, path in enumerate(tqdm(impaths)):
if len(imcaps[i]) < captions_per_image:
captions = imcaps[i] + [choice(imcaps[i]) for _ in range(captions_per_image - len(imcaps[i]))]
else:
captions = sample(imcaps[i], k=captions_per_image)
assert len(captions) == captions_per_image
img = imread(impaths[i])
if len(img.shape) == 2:
img = img[:, :, np.newaxis]
img = np.concatenate([img, img, img], axis=2)
img = imresize(img, (255, 255))
img = img.transpose(2, 0, 1)
assert img.shape == (3, 255, 255)
assert np.max(img) <= 255
images[i] = img
image_paths.append(path)
for j, c in enumerate(captions):
enc_c = [word_map['<start>']] + [word_map.get(word, word_map['<unk>']) for word in c] + [
word_map['<end>']] + [word_map['<pad>']] * (max_len - len(c))
c_len = len(c) + 2
enc_captions.append(enc_c)
caplens.append(c_len)
assert images.shape[0] * captions_per_image == len(enc_captions) == len(caplens) == len(image_paths) * captions_per_image
with open(os.path.join(output_folder, split + '_CAPTIONS_' + base_filename + '.json'), 'w') as j:
json.dump(enc_captions, j)
with open(os.path.join(output_folder, split + '_CAPLENS_' + base_filename + '.json'), 'w') as j:
json.dump(caplens, j)
with open(os.path.join(output_folder, split + '_IMPATHS_' + base_filename + '.json'), 'w') as j:
json.dump(image_paths, j)
def init_embedding(embeddings):
bias = np.sqrt(3.0 / embeddings.size(1))
torch.nn.init.uniform_(embeddings, -bias, bias)
def clip_gradient(optimizer, grad_clip):
for group in optimizer.param_groups:
for param in group['params']:
if param.grad is not None:
param.grad.data.clamp_(-grad_clip, grad_clip)
def save_checkpoint(data_name, epoch, epochs_since_improvement, encoder, decoder, encoder_optimizer, decoder_optimizer,
bleu_list, is_best):
state = {'epoch': epoch,
'epochs_since_improvement': epochs_since_improvement,
'bleu_scores': bleu_list,
'encoder': encoder,
'decoder': decoder,
'encoder_optimizer': encoder_optimizer,
'decoder_optimizer': decoder_optimizer}
filename = 'checkpoint_' + data_name + '.pth.tar'
torch.save(state, filename)
if is_best:
torch.save(state, 'BEST_' + filename)
class AverageMeter(object):
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=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
def adjust_learning_rate(optimizer, shrink_factor):
print("\nDECAYING learning rate.")
for param_group in optimizer.param_groups:
param_group['lr'] = param_group['lr'] * shrink_factor
print("The new learning rate is %f\n" % (optimizer.param_groups[0]['lr'],))
def accuracy(scores, targets, k):
batch_size = targets.size(0)
_, ind = scores.topk(k, 1, True, True)
correct = ind.eq(targets.view(-1, 1).expand_as(ind))
correct_total = correct.view(-1).float().sum() # 0D tensor
return correct_total.item() * (100.0 / batch_size)