-
Notifications
You must be signed in to change notification settings - Fork 2
/
visualize_attention_mechanism.py
181 lines (135 loc) · 6.63 KB
/
visualize_attention_mechanism.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
"""
# Pytorch implementation for TIP2023 paper from
# https://arxiv.org/abs/2303.13371.
# "Plug-and-Play Regulators for Image-Text Matching"
# Haiwen Diao, Ying Zhang, Wei Liu, Xiang Ruan, Huchuan Lu
#
# Writen by Haiwen Diao, 2023
# If you find this code is useful, please cite our paper and star the project. (We do need it! HaHaHaHa.)
# Thanks for the interest in this project.
"""
import os
import cv2 as cv
import nltk
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import numpy as np
import json
from imageio import imread
def visualize_region(img, bboxes, size, attention, save_path, save_name, margin=-1, alpha=0.5):
# the number of regions
k = bboxes.shape[0]
assert k == 36
mask = np.zeros((size['image_h'], size['image_w']))
for i in range(k):
bbox = bboxes[i]
x1, y1, x2, y2 = int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])
if attention[i] > margin:
mask_atten = attention[i] * np.ones((y2-y1, x2-x1))
mask[y1:y2, x1:x2] = np.maximum(mask[y1:y2, x1:x2], mask_atten)
mask = np.uint8((mask[:, :, np.newaxis]).repeat(3, axis=-1) * 255)
# mask = cv.resize(mask, (img.shape[1], img.shape[0]), interpolation=cv.INTER_LINEAR)
heat_img = cv.applyColorMap(mask, cv.COLORMAP_JET)
merge_img = cv.addWeighted(img, alpha, heat_img, 1-alpha, 0)
cv.imwrite(os.path.join(save_path, '{}.png'.format(save_name)), merge_img)
def visualize_word(x, y_value, x_key, save_path, save_name, figsize=[20, 10], labelsize=30):
config = {
"font.family":'serif', # sans-serif/serif/cursive/fantasy/monospace
"font.size": 50, # medium/large/small
'font.style':'normal', # normal/italic/oblique
'font.weight':'normal', # bold
"mathtext.fontset":'cm',# 'cm' (Computer Modern)
"font.serif": ['cmb10'], # 'Simsun'宋体
"axes.unicode_minus": False,# 用来正常显示负号
}
plt.rcParams.update(config)
plt.figure(figsize=figsize)
plt.bar(x, y_value, tick_label=x_key, align='center')
plt.tick_params(axis='x', labelsize=labelsize)
plt.tick_params(axis='y', labelsize=labelsize)
plt.xticks(rotation=30)
plt.show()
plt.savefig(os.path.join(save_path,'{}.png'.format(save_name)))
plt.close()
if __name__ == '__main__':
# If you find this code is useful, please cite our paper and star the project. (We do need it! HaHaHaHa.)
# Thanks for the interest in this project.
# -----------------------initial path settings----------------------------- #
root_image = 'data/f30k/flickr30k-images/'
root_image_id = 'data/f30k/precomp/test_ids.txt'
root_caption = 'data/f30k/precomp/test_caps.txt'
# load from https://github.com/Paranioar/RCAR/tree/main/data/f30k/id_mapping.json
root_image_id_mapping = 'data/f30k/id_mapping.json'
# load from https://drive.google.com/file/d/1ZVLIN7uSh3dqYAEldelyYF2ei9vicJvZ/view
img_bboxes = np.load('data/f30k/precomp/test_ims_bbx.npy')
img_sizes = np.load('data/f30k/precomp/test_ims_size.npy', allow_pickle=True)
caption_list = []
with open(root_caption, 'r') as f:
for line in f:
caption_list.append(line.strip())
with open(root_image_id, 'r') as f:
ids = f.readlines()
image_ids = [int(x.strip()) for x in ids]
with open(root_image_id_mapping, 'r') as f:
id_to_path = json.load(f)
cap_id_list, img_id_list = [], []
for i in range(100):
cap_id_list.append(i)
img_id_list.append(i//5)
model_name = 'name-of-the-saved-model' # need to change #
# ---------------------- initial visual settings--------------------------- #
for cap_id, img_id in zip(cap_id_list, img_id_list):
# load raw word including <start> and <end>
caption = caption_list[cap_id]
print(caption)
tokens = nltk.tokenize.word_tokenize(str(caption).lower())
word_list = []
word_list.append('_start_')
word_list.extend([str(token) for token in tokens])
word_list.append('_end_')
# load raw image, img_bbox, img_size
image = cv.imread(os.path.join(root_image, id_to_path[str(image_ids[img_id*5])]))
# cv.imshow('image', image)
# cv.waitKey(0)
img_bbox = img_bboxes[img_id*5]
img_size = img_sizes[img_id*5]
# ------------------------ word to region -------------------------------- #
file_name = '{}_sentence{}_image{}'.format(model_name, cap_id, img_id)
load_path = 'attention/word2region/weights/'
save_path = 'attention/word2region/images/{}'.format(file_name)
if not os.path.exists(save_path):
os.makedirs(save_path)
attention = np.load('{}{}.npy'.format(load_path, file_name))
print('word2region/caption{}'.format(cap_id), attention.shape)
assert len(word_list) == attention.shape[0]
# shape: (n_word, n_region)
for word_id, word in enumerate(word_list):
print(word)
visualize_region(image, img_bbox, img_size, attention[word_id], save_path,
'{}{}_region_sentence{}_image{}'.format(word_id, word, cap_id, img_id))
# ----------------------- region to word ------------------------------- #
file_name = '{}_sentence{}_image{}'.format(model_name, cap_id, img_id)
load_path = 'attention/region2word/weights/'
save_path = 'attention/region2word/images/{}'.format(file_name)
if not os.path.exists(save_path):
os.makedirs(save_path)
attention = np.load('{}{}.npy'.format(load_path, file_name))
print('region2word/image{}'.format(img_id), attention.shape)
# the number of regions = 36
k = img_bbox.shape[0]
assert k == 36
x = np.arange(0, len(word_list), 1)
x_key = np.array(word_list)
# generate region mask to visual which region
for region_id in range(k):
mask = np.ones((img_size['image_h'], img_size['image_w'])) * 0.3
bbox = img_bbox[region_id]
x1, y1, x2, y2 = int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])
mask[y1:y2, x1:x2] = np.ones((y2 - y1, x2 - x1))
alpha = mask.reshape((img_size['image_h'], img_size['image_w'], 1)) * 255
new_img = np.concatenate((image, alpha), -1)
cv.imwrite(os.path.join(save_path, 'region{}_im{}.png'.format(region_id, img_id)), new_img)
# generate the attention map on sentence words
visualize_word(x, attention[region_id], x_key, save_path,
'{}region_word_sentence{}_image{}'.format(region_id, cap_id, img_id), figsize=[25, 15])