-
Notifications
You must be signed in to change notification settings - Fork 5
/
similarSubArch.py
228 lines (193 loc) · 8.58 KB
/
similarSubArch.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
import torch
import numpy as np
import os
from tools.dataloader import *
import pickle
from PIL import Image
from tools.config import *
import matplotlib.pyplot as plt
import matplotlib as mpl
mpl.use('Agg')
import argparse
parser = argparse.ArgumentParser(description='cam')
parser.add_argument('--model', default='vgg16', type=str, help='model: [vgg16, resnet50, densenet121, darts].')
parser.add_argument('--dataset', default='imagenet', type=str, help='dataset: [imagenet, place].')
parser.add_argument('--epoch', default=20, type=int, help='epoch: from 0 to 20.')
parser.add_argument('--class', default=7, type=int, help='the class wanna to test')
args = parser.parse_args()
if args.dataset == 'imagenet' and args.model == 'vgg16':
from models.imagenet.vgg16 import *
net = vgg16(finding_masks=False).cuda().eval()
if args.dataset == 'imagenet' and args.model == 'resnet50':
from models.imagenet.resnet50 import *
net = resnet50(finding_masks=False).cuda().eval()
if args.dataset == 'imagenet' and args.model == 'densenet121':
from models.imagenet.densenet121 import *
net = densenet121(finding_masks=False).cuda().eval()
if args.dataset == 'imagenet' and args.model == 'darts':
from models.imagenet.darts import *
net = darts(finding_masks=False).cuda().eval()
if args.dataset == 'place' and args.model == 'vgg16':
from models.place.vgg16 import *
net = vgg16(finding_masks=False).cuda().eval()
if args.dataset == 'place' and args.model == 'resnet50':
from models.place.resnet50 import *
net = resnet50(finding_masks=False).cuda().eval()
if args.dataset == 'place' and args.model == 'densenet121':
from models.place.densenet121 import *
net = densenet121(finding_masks=False).cuda().eval()
if args.dataset == 'place' and args.model == 'darts':
from models.place.darts import *
net = darts(finding_masks=False).cuda().eval()
if args.dataset == 'imagenet':
img_dir = get_dataset_dir('imagenet')
class_numbers = 1000
else:
img_dir = get_dataset_dir('place')
class_numbers = 365
total_masks = []
label_list = []
similar_matrix = None
def get_mask(net_file, thresh=0.5):
masks = torch.load(net_file)
ret = []
for layer_name in masks.keys():
ret.append(torch.sigmoid(masks[layer_name].data.squeeze()).gt(thresh).int())
return ret
def compute_similar(i, j):
mask_num = len(total_masks[i])
tot = 0.0
for index in range(mask_num):
tot += float(1.0*(total_masks[i][index]&total_masks[j][index]).sum() / (1.0*(total_masks[i][index]|total_masks[j][index]).sum()))
return tot / (float)(mask_num)
def get_similar_matrix():
print('start calculate similar matrix')
if not os.path.exists('./results'):
os.makedirs('./results')
if not os.path.exists('./results/similarSubArch'):
os.makedirs('./results/similarSubArch')
if not os.path.exists('./results/similarSubArch/{}'.format(args.dataset)):
os.makedirs('./results/similarSubArch/{}'.format(args.dataset))
if not os.path.exists('./results/similarSubArch/{}/{}'.format(args.dataset, args.model)):
os.makedirs('./results/similarSubArch/{}/{}'.format(args.dataset, args.model))
if os.path.exists('./results/similarSubArch/{}/{}/sm.pkl'.format(args.dataset, args.model)):
print('similar matrix has already calculated before')
return
similarity = [[0 for i in range(class_numbers)] for i in range(class_numbers)]
for i in range(class_numbers):
total_masks.append(get_mask('./checkpoint/{}/{}/{:03d}/net_iter{:03d}.pth'.format(args.dataset, args.model, i, args.epoch)))
print('all masks load success.')
for i in range(class_numbers):
if ((i+1) % 100 == 0):
print('calculating class {}...'.format(i+1))
for j in range(i, class_numbers):
similarity[i][j] = similarity[j][i] = compute_similar(i, j)
output = open('./results/similarSubArch/{}/{}/sm.pkl'.format(args.dataset, args.model), 'wb')
pickle.dump(similarity, output)
output.close()
print('similar matrix has been completed')
def load_similar_matrix():
pkl_dir = './results/similarSubArch/{}/{}/sm.pkl'.format(args.dataset, args.model)
pkl_file = open(pkl_dir, 'rb')
similar_matrix = pickle.load(pkl_file)
pkl_file.close()
print('load similar matrix success.')
return similar_matrix
def load_label_list():
if args.dataset == 'imagenet':
label_dir = img_dir + '/synset_words.txt'
elif args.dataset == 'place':
label_dir = img_dir + '/label.txt'
line_count = 0
for line in open(label_dir):
label_list.append('{:03d} {}'.format(line_count, line[:-1]))
line_count += 1
def load_origin_net():
mask_dir = './checkpoint/{}/{}/000/net_iter001.pth'.format(args.dataset, args.model)
net.load_masks(mask_dir)
with torch.no_grad():
masks = net.get_masks()
for key in masks.keys():
mask = (masks[key].data[...])
masks[key].mask[...] = torch.sigmoid(mask).gt(-1)
def get_one_image(class_idx):
val_loader = get_one_image_by_class(image_dir=img_dir, class_num=class_idx, batch_size=1, num_threads=12)
for img, label, img_name in val_loader:
return img, label, img_name
class similar_pair:
def __init__(self, idx, val):
self.idx = idx
self.val = val
def find_top3_similar_subarch(class_idx):
sm_list = []
for i in range(class_numbers):
if class_idx != i:
sm_list.append(similar_pair(i, similar_matrix[class_idx][i]))
sm_list.sort(key=lambda x : (x.val), reverse=True)
return sm_list[:3]
def find_top3_classify(class_idx):
val_loader = get_image_by_class(type='test', image_dir=img_dir, class_num=class_idx, batch_size=1, num_threads=12, crop=224)
for img, _, _ in val_loader:
if torch.cuda.is_available():
img = img.cuda()
if args.model != 'darts':
out = net(img).topk(k=3, dim=1)
else:
out = net(img)[0].topk(k=3, dim=1)
return out
def calculate_and_save_result(class_idx):
if not os.path.exists('./results/similarSubArch/{}/{}/{:03d}'.format(args.dataset, args.model, class_idx)):
os.makedirs('./results/similarSubArch/{}/{}/{:03d}'.format(args.dataset, args.model, class_idx))
f = open('./results/similarSubArch/{}/{}/{:03d}/result.txt'.format(args.dataset, args.model, class_idx), 'w')
f.write('class idx: {:03d}\n'.format(class_idx))
f.write('class name: {}\n'.format(" ".join(label_list[class_idx].split(' ')[2:])))
f.write('\n')
f.write('top3 classify class:\n')
f.write('\n')
res = find_top3_classify(class_idx)
for i in range(3):
f.write('top {:1d} class idx: {:03d}\n'.format(i+1, res[1].squeeze()[i]))
f.write('top {:1d} class name: {}\n'.format(i+1, " ".join(label_list[res[1].squeeze()[i]].split(' ')[2:])))
f.write('top {:1d} class confidence: {:5f}\n'.format(i+1, res[0].squeeze()[i]/100))
f.write('\n')
plt.figure(figsize=(8, 2))
plt.subplot(1, 4, 1)
plt.xticks([])
plt.yticks([])
plt.imshow(get_one_image(class_idx)[0].squeeze().permute(1, 2, 0))
for i in range(3):
plt.subplot(1, 4, i+2)
plt.xticks([])
plt.yticks([])
plt.imshow(get_one_image(res[1].squeeze()[i].cpu().numpy())[0].squeeze().permute(1, 2, 0))
plt.axis('off')
plt.savefig('./results/similarSubArch/{}/{}/{:03d}/classify_classes.png'.format(args.dataset, args.model, class_idx))
plt.close()
f.write('top3 similar subarch:\n')
f.write('\n')
top3_similar_classes = find_top3_similar_subarch(class_idx)
for i in range(3):
f.write('top {:1d} class idx: {:03d}\n'.format(i+1, top3_similar_classes[i].idx))
f.write('top {:1d} class name: {}\n'.format(i+1, " ".join(label_list[top3_similar_classes[i].idx].split(' ')[2:])))
f.write('top {:1d} class similarity: {:5f}\n'.format(i+1, top3_similar_classes[i].val))
f.write('\n')
plt.figure(figsize=(8, 2))
plt.subplot(1, 4, 1)
plt.xticks([])
plt.yticks([])
plt.imshow(get_one_image(class_idx)[0].squeeze().permute(1, 2, 0))
for i in range(3):
plt.subplot(1, 4, i+2)
plt.xticks([])
plt.yticks([])
plt.imshow(get_one_image(top3_similar_classes[i].idx)[0].squeeze().permute(1, 2, 0))
plt.axis('off')
plt.savefig('./results/similarSubArch/{}/{}/{:03d}/similar_subarch.png'.format(args.dataset, args.model, class_idx))
plt.close()
if __name__ == '__main__':
get_similar_matrix()
similar_matrix = load_similar_matrix()
load_label_list()
load_origin_net()
for i in range(100):
calculate_and_save_result(i)