-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmain.py
142 lines (110 loc) · 4.47 KB
/
main.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
from __future__ import print_function
import matplotlib.pyplot as plt
import argparse
import os
import numpy as np
import torch
import torch.optim
import torchvision.models as models
from torchvision import transforms
from torchvision.utils import save_image
import pickle
import random
from tqdm import tqdm
from loss import *
from utils import *
from networks import *
from common import conv
dtype = torch.cuda.FloatTensor
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# TODO: Modify if testing on multiple archs.
pretrained_net = models.resnet18(pretrained=True).to(device)
pretrained_net.eval()
imsize = 227 if pretrained_net == 'alexnet_caffe' else 224
preprocess = get_resnet_preprocessor(imsize)
cnn = pretrained_net.to(device)
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Define hyperparameters.')
# parser.add_argument('--patch_num', type=int, default=2)
parser.add_argument('--patch_num', default='double', nargs='?', choices=['single', 'double', 'repeated', 'null'])
parser.add_argument('--lr', type=float, default=0.01)
parser.add_argument('--num_iter', type=int, default=1500)
parser.add_argument('--indep_thresh', type=float, default=0.14)
parser.add_argument('--accu_thresh', type=float, default=0.94)
parser.add_argument('--var_thresh', type=float, default=0.0001)
parser.add_argument('--save_as_image', type=bool, default=True, help='Save data also as an image.')
parser.add_argument('--dir', type=str, default='./', help='Directory to save data.')
parser.add_argument('--image_num', type=int, default=150, help='Number of images to save.')
args = parser.parse_args()
patch_num = args.patch_num
LR = args.lr
num_iter = args.num_iter
images_saved = 0
imsize_net = 256
plt.style.use('seaborn')
PLOT = False
dir = args.dir
data_dir = os.path.join(dir, 'Data/')
stat_dir = os.path.join(dir,'Stat/')
image_dir = os.path.join(dir, 'Images/')
os.makedirs(dir, exist_ok=True)
os.makedirs(data_dir, exist_ok=True)
os.makedirs(stat_dir, exist_ok=True)
if args.save_as_image:
os.makedirs(image_dir, exist_ok=True)
repeated = False
if patch_num == 'single':
closure_func = closure_single_patch
patch_num = 1
elif patch_num == 'double':
closure_func = closure
patch_num = 2
elif patch_num == 'null':
closure_func = closure_null
patch_num = 2
elif patch_num == 'repeated':
repeated = True
closure_func = closure_repeated_patch
patch_num = 2
while images_saved < args.image_num:
prior_nets = [skip_net() for i in range(patch_num)]
net_inputs = [get_noise(32, 'noise', imsize_net).type(dtype).detach() for i in range(patch_num)]
optimizer = torch.optim.RMSprop([item for i in range(patch_num) for item in prior_nets[i].parameters()], lr=LR)
counter = 0
last_iters = []
map_idxs = random.sample(range(0, 500), 1) + random.sample(range(500, 1000), 1)
print("labels:", map_idxs)
key_p0_t0, key_p0_t1, key_p1_t0, key_p1_t1 = 0.5, 0.5, 0.5, 0.5
alpha, gamma0, gamma1 = 0.1, 0.8, 0.8
for j in tqdm(range(num_iter)):
optimizer.zero_grad()
if args.patch_num == 'single':
keys = key_p0_t0, key_p0_t1
else:
keys = key_p0_t0, key_p0_t1, key_p1_t0, key_p1_t1
p0, p1, pic, ALL, patch_positions, random_bg, keys = closure_func(cnn, preprocess, prior_nets, net_inputs, patch_num, imsize, map_idxs, j, keys)
if args.patch_num == 'single':
key_p0_t0, key_p0_t1 = keys
else:
key_p0_t0, key_p0_t1, key_p1_t0, key_p1_t1 = keys
last_iters += p0
if all(x < args.indep_thresh for x in p1) and j > 50 and all(args.accu_thresh < x for x in p0):
images_saved += 1
f = open( stat_dir + "stat." + str(images_saved) + ".txt","w+")
f.write("indexes: " + str(map_idxs) + "\n")
f.write("p0: " + str(p0) + "\n")
f.write("p1: " + str(p1) + "\n")
f.write("scores: " + " ".join(str(a) for a in ALL))
f.close()
data_file = os.path.join(data_dir, 'data.' + str(images_saved))
f = open(data_file, 'wb')
pickle.dump([patch_positions, random_bg, map_idxs], f)
f.close()
# saving figure
if args.save_as_image:
image_file = os.path.join(image_dir, "image."+str(images_saved)+ "_" +"_".join(str(v) for v in map_idxs)+".pdf")
save_image(pic, image_file)
print("SAVED!")
break;
optimizer.step()
print("Done. Images saved: "+ str(images_saved))