-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtest.py
225 lines (171 loc) · 8.8 KB
/
test.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
import argparse
import os.path as osp
import numpy as np
import torch
from torch import nn, optim, autograd
import torch.nn.functional as F
from torch.utils.data import DataLoader
from model.utils import pprint, set_gpu, ensure_path, AverageMeter, Timer, accuracy, one_hot
from tensorboardX import SummaryWriter
import time
import datetime
from keras.utils import to_categorical
import random
from torch.autograd import Variable
import os
import logging
import sys
'''Train Benchmark'''
def get_args():
parser = argparse.ArgumentParser()
# Basic Parameters
parser.add_argument('--dataset', type=str, default='pacs',
choices=['pacs'])
parser.add_argument('--backbone_class', type=str, default='resnet18', choices=['resnet18'])
# Optimization Parameters
parser.add_argument('--max_epoch', type=int, default=100)
parser.add_argument('--lr', type=float, default=0.05)
parser.add_argument('--init_weights', type=str, default=None)
parser.add_argument('--batch_size', type=int, default=64)
parser.add_argument('--image_size', type=int, default=225)
parser.add_argument('--prefetch', type=int, default=16)
# Model Parameters
parser.add_argument('--model_type', type=str,
choices=['DecAug'], default='DecAug')
parser.add_argument('--balance1', type=float, default=0.01) # the balance parameters for category
parser.add_argument('--balance2', type=float, default=0.01) # the balance parameters for context
parser.add_argument('--balanceorth', type=float, default=0.01) # the balance parameters for orth
parser.add_argument('--perturbation', type=float, default=1)
parser.add_argument('--epsilon', type=float, default=1e-8)
parser.add_argument('--targetdomain', type=str, default='photo', choices=['cartoon', 'art_painting', 'photo', 'sketch'])
parser.add_argument('--pretrain', type=bool, default=True)
# Other Parameters
parser.add_argument('--gpu', default='0')
args, unknown_args = parser.parse_known_args()
set_gpu(args.gpu)
args.concept_path = './saves'
args.init_path = os.path.join('./saves/initialization', 'resnet18.pth')
args.save_path = os.path.join('./exp_log', args.targetdomain)
return args
def get_model(args, s1_data):
if args.model_type == 'DecAug':
from model.models.DecAug import bgor2
model = bgor2(args)
else:
raise ValueError('No Such Model')
if torch.cuda.is_available():
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
model = model.cuda()
return model
def get_optimizer(args, model):
parameters = model.named_parameters()
top_list, bottom_list = [], []
for k, v in parameters:
if 'encoder' in k:
bottom_list.append(v)
else:
top_list.append(v)
if args.warmup > 0:
optimizer_warmup = torch.optim.SGD(top_list,
lr=args.lr, momentum=0.9, nesterov=True, weight_decay=0.0005)
lr_scheduler_warmup = torch.optim.lr_scheduler.LambdaLR(optimizer_warmup, lambda epoch: epoch * (1/args.warmup))
optimizer = torch.optim.SGD(model.parameters(),
lr=args.lr, momentum=0.9, nesterov=True, weight_decay=0.0005)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=args.step, gamma=args.gamma)
return optimizer, lr_scheduler, optimizer_warmup, lr_scheduler_warmup
else:
optimizer = torch.optim.SGD([{'params': top_list, 'lr': args.lr * 10},
{'params':bottom_list}],
lr=args.lr, momentum=0.9, nesterov=True, weight_decay=0.0005)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=args.step, gamma=args.gamma)
return optimizer, lr_scheduler, None, None
def get_loader(args):
if args.dataset == 'pacs':
from model.dataloader.pacsLoader import pacsDataset as Dataset
trainset = Dataset('train', args)
valset = Dataset('val', args)
testset = Dataset('test', args)
else:
raise ValueError('Non-supported Dataset.')
args.num_class = trainset.num_class
args.num_concept = trainset.num_concept
train_loader = DataLoader(dataset=trainset, batch_size=args.batch_size, shuffle=True, num_workers=args.prefetch, pin_memory=True)
val_loader = DataLoader(dataset=valset, batch_size=args.batch_size, shuffle=False, num_workers=args.prefetch, pin_memory=True)
test_loader = DataLoader(dataset=testset, batch_size=args.batch_size, shuffle=False, num_workers=args.prefetch, pin_memory=True)
return train_loader, test_loader, test_loader
def test_model(args, model, test_loader, s1_data):
model.load_state_dict(torch.load(osp.join(args.save_path, 'max_acc.pth'))['params'])
t1 = AverageMeter()
ta = AverageMeter()
tv = AverageMeter()
tc = AverageMeter()
model.eval()
with torch.no_grad():
for i, batch in enumerate(test_loader, 1):
if torch.cuda.is_available():
data, gt_label, gt_concept = [_.cuda() for _ in batch]
else:
data, gt_label, gt_concept = batch[0], batch[1], batch[2]
if args.model_type == 'DecAug':
logits, logits_category, logits_concept, feature, feature_category, feature_concept = model(data)
loss1 = F.cross_entropy(logits_category, gt_label)
loss2 = F.cross_entropy(logits_concept, gt_concept)
parm = {}
for name, parameters in model.named_parameters():
parm[name] = parameters
# concept branch
w_branch = parm['concept_branch.weight']
w_tensor = parm['fcc0.weight']
b_tensor = parm['fcc0.bias']
# classification branch
w_branch_l = parm['category_branch.weight']
w_tensor_l = parm['fc0.weight']
b_tensor_l = parm['fc0.bias']
w_out = parm['classification.weight']
b_out = parm['classification.bias']
w = torch.matmul(w_tensor, w_branch)
grad = -1 * w[gt_concept] + torch.matmul(logits_concept.detach(), w)
grad_norm = grad / (grad.norm(2, dim=1, keepdim=True) + args.epsilon)
w_l = torch.matmul(w_tensor_l, w_branch_l)
grad_l = -1 * w_l[gt_label] + torch.matmul(logits_category.detach(), w_l)
grad_norm_l = grad_l / (grad_l.norm(2, dim=1, keepdim=True) + args.epsilon)
b, L = grad_norm_l.shape
grad_norm = grad_norm.reshape(b, 1, L)
grad_norm_l = grad_norm_l.reshape(b, L, 1)
loss_orth = ((torch.bmm(grad_norm, grad_norm_l).cuda()) ** 2).sum()
grad_aug = -1 * w_tensor[gt_concept] + torch.matmul(logits_concept.detach(), w_tensor)
FGSM_attack = args.perturbation * (grad_aug.detach() / (grad_aug.detach().norm(2, dim=1, keepdim=True) + args.epsilon))
ratio = random.random()
feature_aug = ratio * FGSM_attack
embs = torch.cat((feature_category, feature_concept), 1)
output = torch.matmul(embs, w_out.transpose(0, 1)) + b_out
logits_class = output
loss_class = F.cross_entropy(logits_class, gt_label)
loss = loss_class + args.balance1 * loss1 + args.balance2 * loss2 + args.balanceorth * loss_orth
else:
raise ValueError('')
acc = accuracy(logits_class.data, gt_label.data, topk=(1,))[0]
acc_concept = accuracy(logits_concept.data, gt_concept.data, topk=(1,))[0]
acc_category = accuracy(logits_category.data, gt_label.data, topk=(1,))[0]
t1.update(loss.item(), data.size(0))
ta.update(acc.item(), data.size(0))
tv.update(acc_concept.item(), data.size(0))
tc.update(acc_category.item(), data.size(0))
t1 = t1.avg
ta = ta.avg
tv = tv.avg
tc = tc.avg
logging.info('Test acc={:.4f}, acc_concept={:.4f}, acc_category={:.4f}'.format(ta, tv, tc))
if __name__ == '__main__':
args = get_args()
log_format = '%(asctime)s %(message)s'
logging.basicConfig(stream=sys.stdout, level=logging.INFO,
format=log_format, datefmt='%m/%d %I:%M:%S %p')
fh = logging.FileHandler(os.path.join(args.save_path, 'log.txt'))
fh.setFormatter(logging.Formatter(log_format))
logging.getLogger().addHandler(fh)
logging.info(vars(args))
train_loader, val_loader, test_loader = get_loader(args)
model = get_model(args, logging)
test_model(args, model, test_loader, logging)