-
Notifications
You must be signed in to change notification settings - Fork 0
/
metric.py
178 lines (157 loc) · 7.02 KB
/
metric.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
import os
import cv2
import torch
import numpy as np
import pickle as pkl
import torch.nn as nn
from utils_metric import *
from torchvision import models
# configs
data_dir = './data/CASIA_WebFace_20000'
data_name = 'test_crop'
img_dir = os.path.join(data_dir, data_name)
img_path_list = get_img(img_dir)
id_list = sorted(os.listdir(img_dir))
label_dict = {id_list[i] : i for i in range(50)}
stats_dir = './stats_final/' + 'gradcam'
if not os.path.exists(stats_dir):
os.makedirs(stats_dir)
cam_dir = './result_cam_final/' + 'gradcam'
if not os.path.exists(cam_dir):
os.makedirs(cam_dir)
draw = True
combos = [
['baseModel', 'full', ''], ['baseModel', 'eyes', ''], ['baseModel', 'mouth', ''],
['full', 'eyes', '15'], ['full', 'eyes', '80'],
['full', 'mouth', '15'], ['full', 'mouth', '80'],
['region', 'eyes', '15'], ['region', 'eyes', '80'],
['region', 'mouth', '15'], ['region', 'mouth', '80'],
]
for combo in combos[:1]:
target = combo[0] # 'baseModel' | 'full', 'region'
grp = combo[1] # 'full', 'eyes', 'mouth' | 'eyes', 'mouth'
epoch = combo[2] # | '15', '80'
log_root = './logs/CASIA_WebFace_20000_0.15_final/'
arch = 'resnet50'
lr = 1e-2
gamma = 0.5
bs = 32
# locate log dir
if target == 'baseModel':
if grp == 'full':
train_set, test_set = 'train_crop', 'test_crop'
elif grp == 'eyes':
train_set, test_set = 'train_eyes0.15_crop', 'test_eyes0.15_crop'
else:
train_set, test_set = 'train_mouth0.15_crop', 'test_mouth0.15_crop'
log_dir = log_root + f'[{arch}]_[{lr}]_[{gamma}]_[{bs}]_[{train_set}]_[{test_set}]_[{target}]'
task = f'{grp}_{target}'
elif target == 'full':
train_set, test_set = 'train_crop', 'test_crop'
log_dir = log_root + f'[{arch}]_[{lr}]_[{gamma}]_[{bs}]_[{train_set}]_[{test_set}]_[{grp}0.15_{epoch}]'
task = f'{target}_on_{grp}{epoch}'
else:
if grp == 'eyes':
input = 'mouth'
train_set, test_set = f'train_{input}0.15_crop', f'test_{input}0.15_crop'
else:
input = 'eyes'
train_set, test_set = f'train_{input}0.15_crop', f'test_{input}0.15_crop'
log_dir = log_root + f'[{arch}]_[{lr}]_[{gamma}]_[{bs}]_[{train_set}]_[{test_set}]_[{grp}0.15_{epoch}]'
task = f'{input}_on_{grp}{epoch}'
print(task)
print(log_dir)
# model
model = models.resnet50()
model.fc = nn.Linear(2048, 50)
para_dict = torch.load(os.path.join(log_dir, '150.pth'))
model.load_state_dict(para_dict)
# declare the model used for grad-cam
net = resnet(model)
net.eval()
# storage for res
heatmap_list = []
metrics_list = [] # [Avg_eyes, Avg_mouth, Prop_eyes, Prop_mouth]
eyesRegion_pixel_list = []
mouthRegion_pixel_list = []
# heatmap
for img_path in img_path_list[:]:
cls_name = img_path.split('/')[-2]
img_name = img_path.split('/')[-1].split('.')[0]
print('Now doing:', cls_name + '_' + img_name)
img = read_img(img_path)
pred = net(img)
cls = label_dict[cls_name]
pred[:, cls].backward()
gradients = net.get_activations_gradient()
pooled_gradients = torch.mean(gradients, dim=[0, 2, 3])
activations = net.get_activations(img).detach()
for i in range(len(pooled_gradients)):
activations[:, i, :, :] *= pooled_gradients[i]
heatmap = torch.mean(activations, dim=1).squeeze()
heatmap = np.maximum(heatmap, 0)
heatmap /= torch.max(heatmap)
img = cv2.imread(img_path)
heatmap = cv2.resize(np.array(heatmap), (img.shape[1], img.shape[0]))
heatmap = np.uint8(255 * heatmap)
heatmap_list.append(heatmap)
if draw:
heatmap_color = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)
superimposed_img = heatmap_color * 0.4 + img
save_dir = os.path.join(cam_dir, cls_name + '_' + img_name)
if not os.path.exists(save_dir):
os.makedirs(save_dir)
cv2.imwrite(os.path.join(save_dir, task + '_map.jpg'), superimposed_img)
# load landmarks
landmark_path = img_path.replace('jpg', 'pkl').replace(data_name, data_name.split('_')[0] + '_landmark')
with open(landmark_path, 'rb') as f:
landmark_list = pkl.load(f)[0]
fixes_eyes = [
(landmark_list[0], landmark_list[5]),
(landmark_list[1], landmark_list[6]),
((landmark_list[0] + landmark_list[1]) / 2, (landmark_list[5] + landmark_list[6]) / 2)
]
fixes_mouth = [
(landmark_list[3] , landmark_list[8]),
(landmark_list[4], landmark_list[9]),
((landmark_list[3] + landmark_list[4]) / 2, (landmark_list[8] + landmark_list[9]) / 2 + abs(landmark_list[3] - landmark_list[4])/4)
]
# load bounding box
bbox_path = img_path.replace('jpg','pkl').replace(data_name, data_name.split('_')[0] + '_bbox')
with open(bbox_path, 'rb') as f:
bbox = pkl.load(f)
img_ori_path = img_path.replace(data_name, data_name.split('_')[0])
img_ori = cv2.imread(img_ori_path)
mask_eyes = np.array(get_mask(img_ori, fixes_eyes, bbox))
mask_mouth = np.array(get_mask(img_ori, fixes_mouth, bbox))
eyesRegion_pixel_list.append(np.sum(mask_eyes))
mouthRegion_pixel_list.append(np.sum(mask_mouth))
# print(eyesRegion_pixel_list, mouthRegion_pixel_list)
# draw contour of the metric region
contour_path = os.path.join(save_dir, 'contour.jpg')
if draw and not os.path.exists(contour_path):
img_copy = img.copy()
_, thresh_eyes = cv2.threshold(np.uint8(mask_eyes)*255, 125, 255, cv2.THRESH_BINARY)
contours_eyes, _ = cv2.findContours(thresh_eyes, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
_, thresh_mouth = cv2.threshold(np.uint8(mask_mouth)*255, 125, 255, cv2.THRESH_BINARY)
contours_mouth, _ = cv2.findContours(thresh_mouth, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cv2.drawContours(img_copy, contours_eyes, -1, (0,0,255), 2)
cv2.drawContours(img_copy, contours_mouth, -1, (0,0,255), 2)
cv2.imwrite(contour_path, img_copy)
# compute metric
avg_eyes = zoneIntensityAvg(heatmap, mask_eyes)
avg_mouth = zoneIntensityAvg(heatmap, mask_mouth)
prop_eyes = zoneIntenstityRatio(heatmap, mask_eyes)
prop_mouth = zoneIntenstityRatio(heatmap, mask_mouth)
metrics_list.append([avg_eyes, avg_mouth, prop_eyes, prop_mouth])
# print(metrics_list)
# heatmap_array = np.array(heatmap_list)
# heatmap_path = os.path.join(stats_dir, task + '_heatmaps.pkl')
# with open(heatmap_path, 'wb') as f:
# pkl.dump(np.array(heatmap_array), f)
# metrics_array = np.array(metrics_list)
# metrics_path = os.path.join(stats_dir, task + '_metrics.pkl')
# with open(metrics_path, 'wb') as f:
# pkl.dump(np.array(metrics_array), f)
# print(metrics_array.shape)
# print('saved!')