-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathrecovery.py
130 lines (107 loc) · 5.22 KB
/
recovery.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
from utils import *
from models.classify import *
from models.generator import *
from models.discri import *
import torch
import os
import numpy as np
from attack import inversion, dist_inversion
from argparse import ArgumentParser
torch.manual_seed(9)
parser = ArgumentParser(description='Inversion')
parser.add_argument('--configs', type=str, default='./config/celeba/attacking/ffhq.json')
args = parser.parse_args()
def init_attack_args(cfg):
if cfg["attack"]["method"] =='kedmi':
args.improved_flag = True
args.clipz = True
args.num_seeds = 1
else:
args.improved_flag = False
args.clipz = False
args.num_seeds = 5
if cfg["attack"]["variant"] == 'L_logit' or cfg["attack"]["variant"] == 'ours':
args.loss = 'logit_loss'
else:
args.loss = 'cel'
if cfg["attack"]["variant"] == 'L_aug' or cfg["attack"]["variant"] == 'ours':
args.classid = '0,1,2,3'
else:
args.classid = '0'
if __name__ == "__main__":
# global args, logger
cfg = load_json(json_file=args.configs)
init_attack_args(cfg=cfg)
# Save dir
if args.improved_flag == True:
prefix = os.path.join(cfg["root_path"], "kedmi_300ids")
else:
prefix = os.path.join(cfg["root_path"], "gmi_300ids")
save_folder = os.path.join("{}_{}".format(cfg["dataset"]["name"], cfg["dataset"]["model_name"]), cfg["attack"]["variant"])
prefix = os.path.join(prefix, save_folder)
save_dir = os.path.join(prefix, "latent")
save_img_dir = os.path.join(prefix, "imgs_{}".format(cfg["attack"]["variant"]))
args.log_path = os.path.join(prefix, "invertion_logs")
os.makedirs(prefix, exist_ok=True)
os.makedirs(args.log_path, exist_ok=True)
os.makedirs(save_img_dir, exist_ok=True)
os.makedirs(save_dir, exist_ok=True)
# Load models
targetnets, E, G, D, n_classes, fea_mean, fea_logvar = get_attack_model(args, cfg)
N = 5
bs = 60
# Begin attacking
for i in range(1):
iden = torch.from_numpy(np.arange(bs))
# evaluate on the first 300 identities only
target_cosines = 0
eval_cosines = 0
for idx in range(5):
iden = iden %n_classes
print("--------------------- Attack batch [%s]------------------------------" % idx)
print('Iden:{}'.format(iden))
save_dir_z = '{}/{}_{}'.format(save_dir,i,idx)
if args.improved_flag == True:
#KEDMI
print('kedmi')
dist_inversion(G, D, targetnets, E, iden,
lr=cfg["attack"]["lr"], iter_times=cfg["attack"]["iters_mi"],
momentum=0.9, lamda=100,
clip_range=1, improved=args.improved_flag,
num_seeds=args.num_seeds,
used_loss=args.loss,
prefix=save_dir_z,
save_img_dir=os.path.join(save_img_dir, '{}_'.format(idx)),
fea_mean=fea_mean,
fea_logvar=fea_logvar,
lam=cfg["attack"]["lam"],
clipz=args.clipz)
else:
#GMI
print('gmi')
if cfg["attack"]["same_z"] =='':
inversion(G, D, targetnets, E, iden,
lr=cfg["attack"]["lr"], iter_times=cfg["attack"]["iters_mi"],
momentum=0.9, lamda=100,
clip_range=1, improved=args.improved_flag,
used_loss=args.loss,
prefix=save_dir_z,
save_img_dir=save_img_dir,
num_seeds=args.num_seeds,
fea_mean=fea_mean,
fea_logvar=fea_logvar,lam=cfg["attack"]["lam"],
istart=args.istart)
else:
inversion(G, D, targetnets, E, iden,
lr=args.lr, iter_times=args.iters_mi,
momentum=0.9, lamda=100,
clip_range=1, improved=args.improved_flag,
used_loss=args.loss,
prefix=save_dir_z,
save_img_dir=save_img_dir,
num_seeds=args.num_seeds,
fea_mean=fea_mean,
fea_logvar=fea_logvar,lam=cfg["attack"]["lam"],
istart=args.istart,
same_z='{}/{}_{}'.format(args.same_z,i,idx))
iden = iden + bs