-
Notifications
You must be signed in to change notification settings - Fork 13
/
grid_downsample.py
235 lines (209 loc) · 12.7 KB
/
grid_downsample.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
226
227
228
229
230
231
232
233
234
235
import os
os.environ['OMP_NUM_THREADS'] = '1'
import argparse
import sys
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import datasets
import color_distillation.utils.transforms as T
from color_distillation import models
from color_distillation.trainer import CNNTrainer
from color_distillation.utils.load_checkpoint import checkpoint_loader
from color_distillation.utils.draw_curve import draw_curve
from color_distillation.utils.logger import Logger
from color_distillation.utils.buffer_size_counter import BufferSizeCounter
from color_distillation.utils.image_utils import img_color_denormalize
def main():
# settings
parser = argparse.ArgumentParser(description='Grid-wise down sample')
parser.add_argument('--num_colors', type=int, default=None, help='down sample ratio for area')
parser.add_argument('--sample_type', type=str, default=None,
choices=['mcut', 'octree', 'kmeans', 'jpeg'])
parser.add_argument('--dither', action='store_true', default=False)
parser.add_argument('--jpeg_ratio', type=int, default=None)
parser.add_argument('--train', action='store_true', default=False)
parser.add_argument('-d', '--dataset', type=str, default='cifar10',
choices=['cifar10', 'cifar100', 'stl10', 'svhn', 'imagenet', 'tiny200'])
parser.add_argument('-a', '--arch', type=str, default='vgg16', choices=models.names())
parser.add_argument('-j', '--num_workers', type=int, default=4)
parser.add_argument('-b', '--batch_size', type=int, default=128, metavar='N',
help='input batch size for training (default: 128)')
parser.add_argument('--epochs', type=int, default=60, metavar='N', help='number of epochs to train (default: 10)')
parser.add_argument('--lr', type=float, default=0.1, metavar='LR', help='learning rate (default: 0.1)')
parser.add_argument('--step_size', type=int, default=40)
parser.add_argument('--weight_decay', type=float, default=5e-4)
parser.add_argument('--momentum', type=float, default=0.5, metavar='M', help='SGD momentum (default: 0.5)')
parser.add_argument('--log_interval', type=int, default=100, metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--visualize', action='store_true')
parser.add_argument('--seed', type=int, default=None, help='random seed (default: None)')
args = parser.parse_args()
# seed
if args.seed is not None:
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
else:
torch.backends.cudnn.benchmark = True
if args.dataset == 'svhn' or args.dataset == 'cifar10' or args.dataset == 'cifar100':
H, W, C = 32, 32, 3
elif args.dataset == 'imagenet':
H, W, C = 224, 224, 3
elif args.dataset == 'stl10':
H, W, C = 96, 96, 3
elif args.dataset == 'tiny200':
H, W, C = 64, 64, 3
else:
raise Exception
buffer_size_counter = BufferSizeCounter()
og_trans = [T.PNGCompression(buffer_size_counter)]
if args.sample_type == 'mcut':
sample_trans = [T.MedianCut(args.num_colors, args.dither), T.PNGCompression(buffer_size_counter)]
if args.dither: args.sample_type += '_dither'
elif args.sample_type == 'octree':
sample_trans = [T.OCTree(args.num_colors, args.dither), T.PNGCompression(buffer_size_counter)]
if args.dither: args.sample_type += '_dither'
elif args.sample_type == 'kmeans':
sample_trans = [T.KMeans(args.num_colors, args.dither), T.PNGCompression(buffer_size_counter)]
if args.dither: args.sample_type += '_dither'
elif args.sample_type == 'jpeg':
sample_trans = [T.JpegCompression(buffer_size_counter, args.jpeg_ratio)]
elif args.sample_type is None:
sample_trans = [T.PNGCompression(buffer_size_counter)]
args.sample_type = 'og_img'
else:
raise Exception
# dataset
data_path = os.path.expanduser('~/Data/') + args.dataset
if args.dataset == 'svhn':
num_class = 10
normalize = T.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
sampled_train_trans = T.Compose(sample_trans + [T.ToTensor(), normalize, ])
og_test_trans = T.Compose(og_trans + [T.ToTensor(), normalize, ])
sampled_test_trans = T.Compose(sample_trans + [T.ToTensor(), normalize, ])
denormalizer = img_color_denormalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
sampled_train_set = datasets.SVHN(data_path, split='train', download=True, transform=sampled_train_trans)
og_test_set = datasets.SVHN(data_path, split='test', download=True, transform=og_test_trans)
sampled_test_set = datasets.SVHN(data_path, split='test', download=True, transform=sampled_test_trans)
elif args.dataset == 'cifar10' or args.dataset == 'cifar100':
num_class = 10 if args.dataset == 'cifar10' else 100
normalize = T.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
sampled_train_trans = T.Compose(sample_trans + [T.RandomCrop(32, padding=4),
T.RandomHorizontalFlip(), T.ToTensor(), normalize, ])
og_test_trans = T.Compose(og_trans + [T.ToTensor(), normalize, ])
sampled_test_trans = T.Compose(sample_trans + [T.ToTensor(), normalize, ])
denormalizer = img_color_denormalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
if args.dataset == 'cifar10':
sampled_train_set = datasets.CIFAR10(data_path, train=True, download=True, transform=sampled_train_trans)
og_test_set = datasets.CIFAR10(data_path, train=False, download=True, transform=og_test_trans)
sampled_test_set = datasets.CIFAR10(data_path, train=False, download=True, transform=sampled_test_trans)
else:
sampled_train_set = datasets.CIFAR100(data_path, train=True, download=True, transform=sampled_train_trans)
og_test_set = datasets.CIFAR100(data_path, train=False, download=True, transform=og_test_trans)
sampled_test_set = datasets.CIFAR100(data_path, train=False, download=True, transform=sampled_test_trans)
elif args.dataset == 'imagenet':
num_class = 1000
normalize = T.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
sampled_train_trans = T.Compose(sample_trans + [T.RandomResizedCrop(224), T.RandomHorizontalFlip(),
T.ToTensor(), normalize, ])
og_test_trans = T.Compose(og_trans + [T.Resize(256), T.CenterCrop(224), T.ToTensor(), normalize, ])
sampled_test_trans = T.Compose(sample_trans + [T.Resize(256), T.CenterCrop(224), T.ToTensor(), normalize, ])
denormalizer = img_color_denormalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
sampled_train_set = datasets.ImageNet(data_path, split='train', transform=sampled_train_trans, )
og_test_set = datasets.ImageNet(data_path, split='val', transform=og_test_trans)
sampled_test_set = datasets.ImageNet(data_path, split='val', transform=sampled_test_trans, )
elif args.dataset == 'stl10':
num_class = 10
# smaller batch size
args.batch_size = 32
normalize = T.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
sampled_train_trans = T.Compose(sample_trans + [T.RandomCrop(96, padding=12),
T.RandomHorizontalFlip(), T.ToTensor(), normalize, ])
og_test_trans = T.Compose(og_trans + [T.ToTensor(), normalize, ])
sampled_test_trans = T.Compose(sample_trans + [T.ToTensor(), normalize, ])
denormalizer = img_color_denormalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
sampled_train_set = datasets.STL10(data_path, split='train', download=True, transform=sampled_train_trans)
og_test_set = datasets.STL10(data_path, split='test', download=True, transform=og_test_trans)
sampled_test_set = datasets.STL10(data_path, split='test', download=True, transform=sampled_test_trans)
elif args.dataset == 'tiny200':
num_class = 200
normalize = T.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
sampled_train_trans = T.Compose(sample_trans + [T.RandomCrop(64, padding=8), T.RandomHorizontalFlip(),
T.ToTensor(), normalize, ])
og_test_trans = T.Compose(og_trans + [T.ToTensor(), normalize, ])
sampled_test_trans = T.Compose(sample_trans + [T.ToTensor(), normalize, ])
denormalizer = img_color_denormalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
sampled_train_set = datasets.ImageFolder(data_path + '/train', transform=sampled_train_trans, )
og_test_set = datasets.ImageFolder(data_path + '/val', transform=og_test_trans)
sampled_test_set = datasets.ImageFolder(data_path + '/val', transform=sampled_test_trans, )
else:
raise Exception
sampled_train_loader = torch.utils.data.DataLoader(sampled_train_set, batch_size=args.batch_size, shuffle=True,
num_workers=args.num_workers, pin_memory=True)
og_test_loader = torch.utils.data.DataLoader(og_test_set, batch_size=args.batch_size, shuffle=False,
num_workers=args.num_workers, pin_memory=True)
sampled_test_loader = torch.utils.data.DataLoader(sampled_test_set, batch_size=args.batch_size, shuffle=False,
num_workers=args.num_workers, pin_memory=True)
logdir = 'logs/grid/{}/{}/{}colors'.format(args.dataset, args.arch,
'full_' if args.num_colors is None else args.num_colors)
if args.train:
os.makedirs(logdir, exist_ok=True)
sys.stdout = Logger(os.path.join(logdir, 'log.txt'), )
print('Settings:')
print(vars(args))
# model
model = models.create(args.arch, num_class, not args.train).cuda()
optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum, weight_decay=args.weight_decay)
scheduler = torch.optim.lr_scheduler.OneCycleLR(optimizer, max_lr=args.lr,
steps_per_epoch=len(sampled_train_loader), epochs=args.epochs)
# draw curve
x_epoch = []
train_loss_s = []
train_prec_s = []
og_test_loss_s = []
og_test_prec_s = []
masked_test_loss_s = []
masked_test_prec_s = []
trainer = CNNTrainer(model, nn.CrossEntropyLoss(), args.num_colors,
denormalizer=denormalizer, sample_method=args.sample_type)
# learn
if args.train:
for epoch in range(1, args.epochs + 1):
print('Train on sampled dateset...')
train_loss, train_prec = trainer.train(epoch, sampled_train_loader, optimizer, args.log_interval, scheduler)
print('Test on original dateset...')
og_test_loss, og_test_prec = trainer.test(og_test_loader)
print('Test on sampled dateset...')
masked_test_loss, masked_test_prec = trainer.test(sampled_test_loader)
x_epoch.append(epoch)
train_loss_s.append(train_loss)
train_prec_s.append(train_prec)
og_test_loss_s.append(og_test_loss)
og_test_prec_s.append(og_test_prec)
masked_test_loss_s.append(masked_test_loss)
masked_test_prec_s.append(masked_test_prec)
draw_curve(os.path.join(logdir, 'learning_curve.jpg'), x_epoch, train_loss_s, train_prec_s,
og_test_loss_s, og_test_prec_s, masked_test_loss_s, masked_test_prec_s)
# save
torch.save(model.state_dict(), os.path.join(logdir, 'model.pth'))
else:
if args.dataset != 'imagenet':
resume_dir = 'logs/grid/{}/{}/full_colors'.format(args.dataset, args.arch)
resume_fname = resume_dir + '/model.pth'
# model = checkpoint_loader(model,resume_fname)
model.load_state_dict(torch.load(resume_fname))
model.eval()
# print('Test on original dateset...')
# trainer.test(og_test_loader)
# print(f'Average image size: {buffer_size_counter.size / len(sampled_test_set):.1f}; '
# f'Bit per pixel: {buffer_size_counter.size / len(sampled_test_set) / H / W:.3f}')
buffer_size_counter.reset()
print('Test on sampled dateset...')
trainer.test(sampled_test_loader, args.visualize)
print(f'Average image size: {buffer_size_counter.size / len(sampled_test_set):.1f}; '
f'Bit per pixel: {buffer_size_counter.size / len(sampled_test_set) / H / W:.3f}')
if __name__ == '__main__':
main()