-
Notifications
You must be signed in to change notification settings - Fork 0
/
label_smooth_cifar10.py
169 lines (152 loc) · 9.09 KB
/
label_smooth_cifar10.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
import torch
import random
import torch.nn.functional as F
from typing import OrderedDict
from recovering import label_recovery
from skimage.metrics import structural_similarity as SSIM
from skimage.metrics import peak_signal_noise_ratio as PSNR
import numpy as np
import os
from distutils.util import strtobool
from datetime import datetime
from lpips import LPIPS,im2tensor
import argparse
## train: tv=1e-3 untrain: tv=1e-2
parser = argparse.ArgumentParser(description='setting for image recovery')
parser.add_argument('--seed',default=2023,type=int)
parser.add_argument('--pretrained',default='False',type=str)
parser.add_argument('--costfn',default='sim')
# parser.add_argument('--augmentation',default='label_smooth')
parser.add_argument('--verble',default='false',type=str)
parser.add_argument('--cuda',default='0',type=str)
parser.add_argument('--tv',default=1e-2,type=float)
args = parser.parse_args()
seed=args.seed
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.enabled = False
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
epoch=10
repetition=10
sample_per_class=3
comparison=5
cost_fn=args.costfn
def to_tensor(image):
return im2tensor(np.array(image))
if cost_fn=='l2':
iteration=300
lr=1
optim_fn='lbfgs'
verble=50
lr_decay=True
total_v=args.tv
elif cost_fn=='sim':
iteration=4800
lr=0.1
optim_fn='adam'
verble=1000
lr_decay=True
total_v=args.tv
CONFIG=OrderedDict(device=torch.device('cuda:'+args.cuda),
dataset="cifar10",
network="lenet",
opt="lbfgs",
type='label_smooth',
pretrained=bool(strtobool(args.pretrained)),
lr=0.5,
bound=100,
iteration=200,
initia=1.,
coefficient=4.)
dir_name='data/fc_recovery/final_opt/'+CONFIG['type']+'_'+str(args.tv)+'_cuda:'+args.cuda+'_'+cost_fn+'_'+str(CONFIG['pretrained'])+datetime.strftime(datetime.now(),'%Y-%m-%d %H:%M:%S')
test=label_recovery(CONFIG)
data_index_list=np.load("additional_files/mixup_list_cifar10.npy")
loss_fn_alex = LPIPS(net='alex')
loss_fn_vgg = LPIPS(net='vgg')
rec_label=np.zeros((epoch,sample_per_class,repetition,comparison,test.classes))
prob_list=np.zeros((epoch,sample_per_class))
psnr=np.zeros((epoch,sample_per_class,repetition,comparison))
ssim=np.zeros((epoch,sample_per_class,repetition,comparison))
lpips_alex=np.zeros((epoch,sample_per_class,repetition,comparison))
lpips_vgg=np.zeros((epoch,sample_per_class,repetition,comparison))
image_buffer=torch.zeros((epoch,sample_per_class,repetition,comparison,3,test.size[0],test.size[1]))
image_gt=torch.zeros((epoch,sample_per_class,3,test.size[0],test.size[1]))
runningloss=np.zeros((epoch,sample_per_class,repetition,comparison))
image_index=np.zeros((epoch,sample_per_class))
for i in range(epoch):
image_index[i]=np.random.choice(data_index_list[i],3,replace=False)
if strtobool(args.verble):
os.makedirs(dir_name)
np.save(dir_name+"/image_index.npy",image_index)
for i in range(epoch):
print(f"epoch={i}")
for ii in range(sample_per_class):
print(f"sample {ii}!")
if hasattr(test,"recover_label"):
del test.recover_label
while not hasattr(test,"recover_label"):
prob=random.uniform(0,0.5)
test.setup(int(image_index[i,ii]),prob)
image_gt[i,ii]=test.origin_data[0].cpu()
test.label_reco()
if not hasattr(test,"recover_label"):
test.pso()
prob_list[i,ii]=prob
test.create_opt_dlg_label()
if strtobool(args.verble):
np.save(dir_name+'/prob_list.npy',prob_list)
torch.save(image_gt,dir_name+"/image_gt.pt")
for time in range(repetition):
print(f'repetition {time}!')
test.dummy_image=torch.randn(test.origin_data.size())
test.reconstruct(iteration=iteration, cost_fn=cost_fn, lr=lr, optim_fn=optim_fn, magnify=1,label='origin',verble=verble,lr_decay=lr_decay,total_variation=total_v,keep=False,record_picking=True)
image_buffer[i,ii,time,0]=test.dummy_data.detach().cpu()[0]
psnr[i,ii,time,0]=PSNR(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),data_range=256)
ssim[i,ii,time,0]=SSIM(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),channel_axis=2)
lpips_alex[i,ii,time,0]=loss_fn_alex.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
lpips_vgg[i,ii,time,0]=loss_fn_vgg.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
runningloss[i,ii,time,0]=test.runningloss
rec_label[i,ii,time,0]=np.array(test.dummy_label.detach().cpu())
test.reconstruct(iteration=iteration, cost_fn=cost_fn, lr=lr, optim_fn=optim_fn, magnify=1,label='hard',verble=verble,lr_decay=lr_decay,total_variation=total_v,keep=False,record_picking=True)
image_buffer[i,ii,time,1]=test.dummy_data.detach().cpu()[0]
psnr[i,ii,time,1]=PSNR(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),data_range=256)
ssim[i,ii,time,1]=SSIM(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),channel_axis=2)
lpips_alex[i,ii,time,1]=loss_fn_alex.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
lpips_vgg[i,ii,time,1]=loss_fn_vgg.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
runningloss[i,ii,time,1]=test.runningloss
rec_label[i,ii,time,1]=np.array(test.dummy_label.detach().cpu())
test.reconstruct(iteration=iteration, cost_fn=cost_fn, lr=lr, optim_fn=optim_fn, magnify=1,label='dlg',verble=verble,lr_decay=lr_decay,total_variation=total_v,keep=False,record_picking=True)
image_buffer[i,ii,time,2]=test.dummy_data.detach().cpu()[0]
psnr[i,ii,time,2]=PSNR(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),data_range=256)
ssim[i,ii,time,2]=SSIM(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),channel_axis=2)
lpips_alex[i,ii,time,2]=loss_fn_alex.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
lpips_vgg[i,ii,time,2]=loss_fn_vgg.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
runningloss[i,ii,time,2]=test.runningloss
rec_label[i,ii,time,2]=np.array(F.softmax(test.dummy_label.detach().cpu(), dim=-1))
test.reconstruct(iteration=iteration, cost_fn=cost_fn, lr=lr, optim_fn=optim_fn, magnify=1,label='optimal',verble=verble,lr_decay=lr_decay,total_variation=total_v,keep=False,record_picking=True)
rec_label[i,ii,time,3]=np.array(test.dummy_label.detach().cpu())
image_buffer[i,ii,time,3]=test.dummy_data.detach().cpu()[0]
psnr[i,ii,time,3]=PSNR(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),data_range=256)
ssim[i,ii,time,3]=SSIM(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),channel_axis=2)
lpips_alex[i,ii,time,3]=loss_fn_alex.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
lpips_vgg[i,ii,time,3]=loss_fn_vgg.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
runningloss[i,ii,time,3]=test.runningloss
test.reconstruct(iteration=iteration, cost_fn=cost_fn, lr=lr, optim_fn=optim_fn, magnify=1,label='opt_dlg',verble=verble,lr_decay=lr_decay,total_variation=total_v,keep=False,record_picking=True)
rec_label[i,ii,time,4]=np.array(test.dummy_label.detach().cpu())
image_buffer[i,ii,time,4]=test.dummy_data.detach().cpu()[0]
psnr[i,ii,time,4]=PSNR(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),data_range=256)
ssim[i,ii,time,4]=SSIM(np.array(test.tp(test.origin_data[0].cpu())), np.array(test.tp(test.dummy_data[0].cpu())),channel_axis=2)
lpips_alex[i,ii,time,4]=loss_fn_alex.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
lpips_vgg[i,ii,time,4]=loss_fn_vgg.forward(to_tensor(test.tp(test.origin_data[0].cpu())),to_tensor(test.tp(test.dummy_data[0].cpu())))
runningloss[i,ii,time,4]=test.runningloss
if strtobool(args.verble):
torch.save(image_buffer,dir_name+"/image_buffer.pt")
np.save(dir_name+"/psnr.npy",psnr)
np.save(dir_name+"/ssim.npy",ssim)
np.save(dir_name+"/lpips_alex.npy",lpips_alex)
np.save(dir_name+"/lpips_vgg.npy",lpips_vgg)
np.save(dir_name+"/runningloss.npy",runningloss)
np.save(dir_name+"/rec_label.npy",rec_label)