-
Notifications
You must be signed in to change notification settings - Fork 3
/
gene_pairs.py
127 lines (105 loc) · 4.88 KB
/
gene_pairs.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
import sys
import sys, os
import argparse
import time
import copy
import numpy as np
def gene_local_points(grid_size, pair_num, local_size=1):
neighbor = []
for j in range(-local_size, local_size + 1):
for h in range(-local_size, local_size + 1):
if j == 0 and h == 0:
pass
else:
item = [j, h]
neighbor.append(item)
neighbor = np.array(neighbor)
tot_pairs = []
for k in range(pair_num):
while True:
x1 = np.random.randint(0, grid_size)
y1 = np.random.randint(0, grid_size)
point1 = x1 * grid_size + y1
np.random.shuffle(neighbor)
x2 = x1 + neighbor[0][0]
y2 = y1 + neighbor[0][1]
if x2 < 0:
x2 = 0
elif x2 >= grid_size:
x2 = grid_size - 1
if y2 < 0:
y2 = 0
elif y2 >= grid_size:
y2 = grid_size - 1
point2 = x2 * grid_size + y2
if point1 == point2: # bugFix
continue
if [point1, point2] in tot_pairs or [point2, point1] in tot_pairs:
continue
else:
break
tot_pairs.append(list([point1, point2]))
tot_pairs = np.array(tot_pairs)
return tot_pairs
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--seed", default=0, type=int, help="random seed")
parser.add_argument("--ratios", default=[0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.85,0.9,0.95,1], type=list,
help='ratios of context')
parser.add_argument("--sample_num", default=100, type=int, help='sample num of S')
parser.add_argument("--grid_size", default=16, type=int, help='number of grids of each img')
parser.add_argument("--pair_num", default=200, type=int, help='number of point pair of each test img')
parser.add_argument("--targeted", action="store_true", dest="targeted", help="whether use the targeted attack (True for targeted attack, False for untargeted attack)")
# prefix of save path
parser.add_argument('--distance', default='l_inf', type=str,help="type of adversarial attacks, currently only support 'l_inf'")
parser.add_argument('--arch', default="resnet18", type=str, help="model name")
parser.add_argument("--adv_model", action="store_true", dest="adv_model", help="the type of model (True for adversarially learned DNN, False for standardly learned DNN)")
args = parser.parse_args()
np.random.seed(args.seed)
if not args.targeted:
args.img_adv = "advImgs_untarget"
prefix = "{}/{}/untarget/".format(args.distance, args.arch)
else:
args.img_adv = "advImgs_target"
prefix = "{}/{}/target/".format(args.distance, args.arch)
if args.adv_model:
args.selected_imgs = os.listdir(os.path.join(prefix + args.img_adv, "adv_model"))
args.point_path = os.path.join(args.img_adv, "adv_model", "points")
else:
args.selected_imgs = os.listdir(os.path.join(prefix + args.img_adv, "ori_model"))
args.point_path = os.path.join(args.img_adv, "ori_model", "points")
args.selected_imgs.sort()
point_dir = os.path.join(prefix, args.point_path)
if not os.path.exists(point_dir):
os.makedirs(point_dir)
# gene pairs for each image
for im in args.selected_imgs:
if not im.startswith('ILSVRC'):
continue
img_name = im.replace('.npy', '')
print('Image ', img_name)
save_path = os.path.join(point_dir, "img{}".format(img_name))
if not os.path.exists(save_path):
os.makedirs(save_path)
tot_pairs = gene_local_points(args.grid_size, args.pair_num, local_size=1)
np.save(save_path + '/points.npy', tot_pairs)
# tot_pairs = np.load(save_path + '/points.npy')
for r in args.ratios:
print('Ratio:', r)
players = []
for p, pt in enumerate(tot_pairs):
point1, point2 = pt[0], pt[1]
# m-order interactions
context = list(range(args.grid_size ** 2))
context.remove(point1)
context.remove(point2)
players_thispair = []
m = int((args.grid_size ** 2 - 2) * r) # m-order
for k in range(args.sample_num):
players_thispair.append(np.random.choice(context, m, replace=False))
players.append(players_thispair)
player_save_path = os.path.join(point_dir, "img{}/ratio{}_S.npy".format(img_name, int(r * 100)))
players = np.array(players)
np.save(player_save_path, players)
if __name__ == "__main__":
main()