forked from YU1ut/imprinted-weights
-
Notifications
You must be signed in to change notification settings - Fork 0
/
imprint.py
202 lines (169 loc) · 7.13 KB
/
imprint.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
import argparse
import os
import shutil
import time
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.distributed as dist
import torch.optim
import torch.utils.data
import torchvision.transforms as transforms
import models
import loader
from utils import Bar, Logger, AverageMeter, accuracy, mkdir_p, savefig
parser = argparse.ArgumentParser(description='PyTorch ImageNet Training')
parser.add_argument('--data', metavar='DIR', default='CUB_200_2011',
help='path to dataset')
parser.add_argument('-j', '--workers', default=4, type=int, metavar='N',
help='number of data loading workers (default: 4)')
parser.add_argument('-b', '--batch-size', default=64, type=int,
metavar='N', help='mini-batch size (default: 64)')
parser.add_argument('-c', '--checkpoint', default='imprint_checkpoint', type=str, metavar='PATH',
help='path to save checkpoint (default: imprint_checkpoint)')
parser.add_argument('--model', default='', type=str, metavar='PATH',
help='path to model (default: none)')
parser.add_argument('--random', action='store_true', help='whether use random novel weights')
parser.add_argument('--num-sample', default=1, type=int,
metavar='N', help='number of novel sample (default: 1)')
parser.add_argument('--test-novel-only', action='store_true', help='whether only test on novel classes')
parser.add_argument('--aug', action='store_true', help='whether use data augmentation during training')
best_prec1 = 0
def main():
global args, best_prec1
args = parser.parse_args()
if not os.path.isdir(args.checkpoint):
mkdir_p(args.checkpoint)
model = models.Net().cuda()
print('==> Reading from model checkpoint..')
assert os.path.isfile(args.model), 'Error: no model checkpoint directory found!'
checkpoint = torch.load(args.model)
args.start_epoch = checkpoint['epoch']
best_prec1 = checkpoint['best_prec1']
model.load_state_dict(checkpoint['state_dict'])
print("=> loaded model checkpoint '{}' (epoch {})"
.format(args.model, checkpoint['epoch']))
cudnn.benchmark = True
# Data loading code
normalize = transforms.Normalize(mean=[0.5, 0.5, 0.5],
std=[0.5, 0.5, 0.5])
novel_trasforms = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize]) if not args.aug else transforms.Compose([
transforms.Resize(256),
transforms.RandomCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
normalize,
])
novel_dataset = loader.ImageLoader(
args.data,
novel_trasforms,
train=True, num_classes=200,
num_train_sample=args.num_sample,
novel_only=True, aug=args.aug)
novel_loader = torch.utils.data.DataLoader(
novel_dataset, batch_size=args.batch_size, shuffle=False,
num_workers=args.workers, pin_memory=True)
val_loader = torch.utils.data.DataLoader(
loader.ImageLoader(args.data, transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
]), num_classes=200, novel_only=args.test_novel_only),
batch_size=args.batch_size, shuffle=False,
num_workers=args.workers, pin_memory=True)
imprint(novel_loader, model)
test_acc = validate(val_loader, model)
save_checkpoint({
'state_dict': model.state_dict(),
'best_prec1': test_acc,
}, checkpoint=args.checkpoint)
def imprint(novel_loader, model):
batch_time = AverageMeter()
data_time = AverageMeter()
# switch to evaluate mode
model.eval()
end = time.time()
bar = Bar('Imprinting', max=len(novel_loader))
with torch.no_grad():
for batch_idx, (input, target) in enumerate(novel_loader):
# measure data loading time
data_time.update(time.time() - end)
input = input.cuda()
# compute output
output = model.extract(input)
if batch_idx == 0:
output_stack = output
target_stack = target
else:
output_stack = torch.cat((output_stack, output), 0)
target_stack = torch.cat((target_stack, target), 0)
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
# plot progress
bar.suffix = '({batch}/{size}) Data: {data:.3f}s | Batch: {bt:.3f}s | Total: {total:} | ETA: {eta:}'.format(
batch=batch_idx + 1,
size=len(novel_loader),
data=data_time.val,
bt=batch_time.val,
total=bar.elapsed_td,
eta=bar.eta_td
)
bar.next()
bar.finish()
new_weight = torch.zeros(100, 256)
for i in range(100):
tmp = output_stack[target_stack == (i + 100)].mean(0) if not args.random else torch.randn(256)
new_weight[i] = tmp / tmp.norm(p=2)
weight = torch.cat((model.classifier.fc.weight.data, new_weight.cuda()))
model.classifier.fc = nn.Linear(256, 200, bias=False)
model.classifier.fc.weight.data = weight
def validate(val_loader, model):
batch_time = AverageMeter()
data_time = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
# switch to evaluate mode
model.eval()
bar = Bar('Testing ', max=len(val_loader))
with torch.no_grad():
end = time.time()
for batch_idx, (input, target) in enumerate(val_loader):
# measure data loading time
data_time.update(time.time() - end)
input = input.cuda()
target = target.cuda(non_blocking=True)
# compute output
output = model(input)
# measure accuracy
prec1, prec5 = accuracy(output, target, topk=(1, 5))
top1.update(prec1.item(), input.size(0))
top5.update(prec5.item(), input.size(0))
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
# plot progress
bar.suffix = '({batch}/{size}) Data: {data:.3f}s | Batch: {bt:.3f}s | Total: {total:} | ETA: {eta:} | top1: {top1: .4f} | top5: {top5: .4f}'.format(
batch=batch_idx + 1,
size=len(val_loader),
data=data_time.avg,
bt=batch_time.avg,
total=bar.elapsed_td,
eta=bar.eta_td,
top1=top1.avg,
top5=top5.avg,
)
bar.next()
bar.finish()
return top1.avg
def save_checkpoint(state, checkpoint='checkpoint', filename='checkpoint.pth.tar'):
filepath = os.path.join(checkpoint, filename)
torch.save(state, filepath)
if __name__ == '__main__':
main()