-
Notifications
You must be signed in to change notification settings - Fork 2
/
train_clothing1m_ce.py
171 lines (149 loc) · 8.22 KB
/
train_clothing1m_ce.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
import os
import math
import time
import argparse
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torch.backends.cudnn as cudnn
from torchvision import transforms
from utils import train, test, get_pred
from dataset import Clothing1M
from networks.resnet import resnet50
def log(path, str):
print(str)
with open(path, 'a') as file:
file.write(str)
def main():
# Settings
parser = argparse.ArgumentParser(description='PyTorch Clothing1M')
parser.add_argument('--batch_size', type=int, default=256, help='input batch size for training')
parser.add_argument('--epochs', type=int, default=10, help='number of epochs to train')
parser.add_argument('--lr', type=float, default=1e-3, help='init learning rate')
parser.add_argument('--save_model', action='store_true', default=False, help='For Saving the current Model')
parser.add_argument('--use_noisy_val', action='store_true', default=False, help='Using the noisy validation setting. By default, using the benchmark setting.')
parser.add_argument('--init_path', type=str, default=None, help='Path of a pretrained model)')
parser.add_argument('--teacher_path', type=str, default=None, help='Path of the teacher model')
parser.add_argument('--soft_targets', type=bool, default=True, help='Use soft targets')
parser.add_argument('--n_gpu', type=int, default=2, help='number of gpu to use')
parser.add_argument('--test_batch_size', type=int, default=256, help='input batch size for testing')
parser.add_argument('--root', type=str, default='data/Clothing1M/', help='root of dataset')
parser.add_argument('--seed', type=int, default=0, help='random seed')
args = parser.parse_args()
if args.teacher_path is None:
exp_name = 'clothing1m_batch{}_seed{}'.format(args.batch_size, args.seed)
else:
teacher_name = args.teacher_path.replace('models/', '')
teacher_name = teacher_name[:teacher_name.find('_')]
if 'net1' in args.teacher_path:
teacher_name = teacher_name+'net1'
elif 'net2' in args.teacher_path:
teacher_name = teacher_name+'net2'
if args.soft_targets:
exp_name = 'softstudent_of_{}_clothing1m_batch{}_seed{}'.format(teacher_name, args.batch_size, args.seed)
else:
exp_name = 'student_of_{}_clothing1m_batch{}_seed{}'.format(teacher_name, args.batch_size, args.seed)
if args.init_path is None:
args.init_path = args.teacher_path
if args.use_noisy_val:
exp_name = 'nv_'+exp_name
logpath = '{}.txt'.format(exp_name)
log(logpath, 'Settings: {}\n'.format(args))
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
# soft loss
def soft_cross_entropy(output, target):
output = F.log_softmax(output, dim=1)
loss = -torch.mean(torch.sum(output*target, dim=1))
return loss
# Datasets
root = args.root
num_classes = 14
kwargs = {'num_workers': 32, 'pin_memory': True} if torch.cuda.is_available() else {}
train_transform = transforms.Compose([transforms.Resize((256)),
transforms.RandomCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.6959, 0.6537, 0.6371),(0.3113, 0.3192, 0.3214)),
])
test_transform = transforms.Compose([transforms.Resize((256)),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize((0.6959, 0.6537, 0.6371),(0.3113, 0.3192, 0.3214)),
])
train_dataset = Clothing1M(root, mode='train', transform=train_transform, use_noisy_val=args.use_noisy_val)
val_dataset = Clothing1M(root, mode='val', transform=test_transform, use_noisy_val=args.use_noisy_val)
test_dataset = Clothing1M(root, mode='test', transform=test_transform, use_noisy_val=args.use_noisy_val)
train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True, **kwargs)
val_loader = torch.utils.data.DataLoader(val_dataset, batch_size=args.test_batch_size, shuffle=False, **kwargs)
test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=args.test_batch_size, shuffle=False, **kwargs)
if args.teacher_path is not None:
teacher_model = resnet50(num_classes=num_classes).to(device)
teacher_model = torch.nn.DataParallel(teacher_model, device_ids=list(range(args.n_gpu)))
state_dict = torch.load(args.teacher_path)
if not list(state_dict.keys())[0][:7]=='module.' :
state_dict = dict(('module.'+key, value) for (key, value) in state_dict.items())
teacher_model.load_state_dict(state_dict)
distill_dataset = Clothing1M(root, mode='train', transform=test_transform, use_noisy_val=args.use_noisy_val)
if args.soft_targets:
pred = get_pred(teacher_model, device, distill_dataset, args.test_batch_size, num_workers=32, output_softmax=True)
train_criterion = soft_cross_entropy
else:
pred = get_pred(teacher_model, device, distill_dataset, args.test_batch_size, num_workers=32)
train_criterion = F.cross_entropy
train_dataset.targets = pred
log(logpath, 'Get label from teacher {}.\n'.format(args.teacher_path))
del teacher_model
else:
train_criterion = F.cross_entropy
# Building model
def learning_rate(lr_init, epoch):
optim_factor = 0
if(epoch > 5):
optim_factor = 1
return lr_init*math.pow(0.1, optim_factor)
model = resnet50(pretrained=True)
model.fc = nn.Linear(2048, num_classes)
model = torch.nn.DataParallel(model.to(device), device_ids=list(range(args.n_gpu)))
if args.init_path is not None:
state_dict = torch.load(args.init_path)
if not list(state_dict.keys())[0][:7]=='module.' :
state_dict = dict(('module.'+key, value) for (key, value) in state_dict.items())
model.load_state_dict(state_dict)
_, test_acc = test(args, model, device, test_loader, criterion=F.cross_entropy)
log(logpath, 'Initialized testing accuracy: {:.2f}\n'.format(100*test_acc))
cudnn.benchmark = True # Accelerate training by enabling the inbuilt cudnn auto-tuner to find the best algorithm to use for your hardware.
optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=0.9, weight_decay=1e-3)
# Training
save_every_epoch = True
if save_every_epoch:
vals = []
directory = 'models/'+exp_name
if not os.path.exists(directory):
os.makedirs(directory)
val_best, epoch_best, test_at_best = 0, 0, 0
for epoch in range(1, args.epochs + 1):
t0 = time.time()
lr = learning_rate(args.lr, epoch)
for param_group in optimizer.param_groups:
param_group['lr'] = lr
_, train_acc = train(args, model, device, train_loader, optimizer, epoch, criterion=train_criterion)
_, val_acc = test(args, model, device, val_loader, criterion=F.cross_entropy)
_, test_acc = test(args, model, device, test_loader, criterion=F.cross_entropy)
if val_acc>val_best:
val_best, test_at_best, epoch_best = val_acc, test_acc, epoch
if args.save_model:
torch.save(model.state_dict(), '{}_best.pth'.format(exp_name))
if save_every_epoch:
vals.append(val_acc)
torch.save(model.state_dict(), '{}/epoch{}.pth'.format(directory, epoch))
log(logpath, 'Epoch: {}/{}, Time: {:.1f}s. '.format(epoch, args.epochs, time.time()-t0))
log(logpath, 'Train: {:.2f}%, Val: {:.2f}%, Test: {:.2f}%; Val_best: {:.2f}%, Test_at_best: {:.2f}%, Epoch_best: {}\n'.format(
100*train_acc, 100*val_acc, 100*test_acc, 100*val_best, 100*test_at_best, epoch_best))
if save_every_epoch:
np.save('{}/val.npy'.format(directory), vals)
if __name__ == '__main__':
main()