-
Notifications
You must be signed in to change notification settings - Fork 8
/
main_pretrain.py
315 lines (259 loc) · 12.6 KB
/
main_pretrain.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
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
# Copyright (c) ByteDance, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# MAE:https://github.com/facebookresearch/mae
# --------------------------------------------------------
from engine_pretrain import train_one_epoch
import models_mae
from util.misc import NativeScalerWithGradNormCount as NativeScaler
import util.misc as misc
import timm.optim.optim_factory as optim_factory
import argparse
import datetime
import json
import numpy as np
import os
import time
from pathlib import Path
import shutil
import warnings
import torch
import torch.backends.cudnn as cudnn
from torch.utils.tensorboard import SummaryWriter
import torchvision.transforms as transforms
import torchvision.datasets as datasets
def get_args_parser():
parser = argparse.ArgumentParser('dBOT pre-training', add_help=False)
parser.add_argument('--batch_size', default=64, type=int,
help='Batch size per GPU (effective batch size is batch_size * accum_iter * # gpus')
parser.add_argument('--epochs', default=1600, type=int)
parser.add_argument('--accum_iter', default=1, type=int,
help='Accumulate gradient iterations (for increasing the effective batch size under memory constraints)')
# Model parameters
parser.add_argument('--model', default='mae_vit_large_patch16', type=str, metavar='MODEL',
help='Name of model to train')
parser.add_argument('--input_size', default=224, type=int,
help='images input size')
parser.add_argument('--mask_ratio', default=0.75, type=float,
help='Masking ratio (percentage of removed patches).')
# multi-stage distill parameters
parser.add_argument('--depth', nargs='+', type=int,
help="multi stage encoder depth e.g., 0 12 12 ")
parser.add_argument('--init_teacher_encoder_depth', type=int, default=0,
help='initial encoder depth for teacher, special for multi-structure')
parser.add_argument('--stage_epochs', nargs='+', type=int,
help="multi stage trainging, in ema-epochs, we drop momentum to 0, e.g., 0 800 1600")
parser.add_argument('--tlayernorm', type=int, default=0, choices=[0, 1],
help="0: without teache rlayernorm \
1:with vit original self.norm")
parser.add_argument('--sdpr', type=float, default=0.0,
help="vit drop path rate for student ")
parser.add_argument('--tdpr', type=float, default=0.0,
help="vit drop path rate for teacher")
parser.add_argument('--norm_feature_loss', action='store_true',
help='Use (per-patch) normalized pixels as targets for computing loss')
parser.add_argument('--clip_grad', type=float,
default=None, help="clip grad when backpropogate")
# Optimizer parameters
parser.add_argument('--weight_decay', type=float, default=0.05,
help='weight decay (default: 0.05)')
parser.add_argument('--lr', type=float, default=None, metavar='LR',
help='learning rate (absolute lr)')
parser.add_argument('--blr', type=float, default=1.5e-4, metavar='LR',
help='base learning rate: absolute_lr = base_lr * total_batch_size / 256')
parser.add_argument('--min_lr', type=float, default=0., metavar='LR',
help='lower lr bound for cyclic schedulers that hit 0')
parser.add_argument('--warmup_epochs', type=int, default=40, metavar='N',
help='epochs to warmup LR')
# Dataset parameters
parser.add_argument('--data_path', default='~/data/imagenet', type=str,
help='dataset path')
parser.add_argument('--output_dir', default='./output_dir',
help='path where to save, empty for no saving')
parser.add_argument('--log_dir', default='./output_dir',
help='path where to tensorboard log')
parser.add_argument('--device', default='cuda',
help='device to use for training / testing')
parser.add_argument('--seed', default=0, type=int)
parser.add_argument('--resume', default='',
help='resume from checkpoint')
parser.add_argument('--teacher_resume', default='',
help='teacher resume from checkpoint')
parser.add_argument('--start_epoch', default=0, type=int, metavar='N',
help='start epoch')
# load dataset with num_workers
parser.add_argument('--num_workers', default=10, type=int)
parser.add_argument('--pin_mem', action='store_true',
help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
parser.add_argument('--no_pin_mem', action='store_false', dest='pin_mem')
parser.set_defaults(pin_mem=True)
# distributed training parameters
parser.add_argument('--world_size', default=1, type=int,
help='number of distributed processes')
parser.add_argument('--local_rank', default=-1, type=int)
parser.add_argument('--dist_on_itp', action='store_true')
parser.add_argument('--dist_url', default='env://',
help='url used to set up distributed training')
return parser
def main(args):
misc.init_distributed_mode(args)
print('job dir: {}'.format(os.path.dirname(os.path.realpath(__file__))))
print("{}".format(args).replace(', ', ',\n'))
if misc.is_main_process():
with open(os.path.join(args.output_dir, "log.txt"), mode="a", encoding="utf-8") as f:
for key, value in vars(args).items():
f.write('%s:%s\n' % (key, value))
print(key, value)
device = torch.device(args.device)
# fix the seed for reproducibility
seed = args.seed + misc.get_rank()
torch.manual_seed(seed)
np.random.seed(seed)
cudnn.benchmark = True
transform_train = transforms.Compose([
transforms.RandomResizedCrop(args.input_size, scale=(
0.2, 1.0), interpolation=3), # 3 is bicubic
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[
0.229, 0.224, 0.225])
])
dataset_train = datasets.ImageFolder(os.path.join(args.data_path, 'train'), transform=transform_train)
print(dataset_train)
if True:
num_tasks = misc.get_world_size()
global_rank = misc.get_rank()
sampler_train = torch.utils.data.DistributedSampler(
dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
)
print("Sampler_train = %s" % str(sampler_train))
else:
sampler_train = torch.utils.data.RandomSampler(dataset_train)
if global_rank == 0 and args.log_dir is not None:
os.makedirs(args.log_dir, exist_ok=True)
log_writer = SummaryWriter(log_dir=args.log_dir)
else:
log_writer = None
data_loader_train = torch.utils.data.DataLoader(
dataset_train, sampler=sampler_train,
batch_size=args.batch_size,
num_workers=args.num_workers,
pin_memory=args.pin_mem,
drop_last=True,
)
if args.resume and os.path.exists(args.resume):
misc.load_start_epoch(args)
init_index = misc.find_stage_index(args.start_epoch, args)
args.teacher_encoder_depth = [
args.init_teacher_encoder_depth] + args.depth[:-1]
# define the model
model = models_mae.__dict__[args.model](
norm_feature_loss=args.norm_feature_loss,
depth=args.depth[init_index],
drop_path_rate=args.sdpr)
teacher = models_mae.__dict__[args.model](
depth=args.teacher_encoder_depth[init_index],
decoder_depth=0,
drop_path_rate=args.tdpr)
model.to(device)
teacher.to(device)
model_without_ddp = model
teacher_without_ddp = teacher
if misc.is_main_process():
print("Model = %s" % str(model_without_ddp))
print("teacher = %s" % str(teacher_without_ddp))
eff_batch_size = args.batch_size * args.accum_iter * misc.get_world_size()
if args.lr is None:
args.lr = args.blr * eff_batch_size / 256
if misc.is_main_process():
print("base lr: %.2e" % (args.lr * 256 / eff_batch_size))
print("actual lr: %.2e" % args.lr)
print("accumulate grad iterations: %d" % args.accum_iter)
print("effective batch size: %d" % eff_batch_size)
if args.distributed:
model = torch.nn.parallel.DistributedDataParallel(
model, device_ids=[args.gpu], find_unused_parameters=True)
model_without_ddp = model.module
teacher_without_ddp.load_state_dict(
model.module.state_dict(), strict=False)
for param in teacher.parameters():
param.requires_grad = False
param_groups = optim_factory.add_weight_decay(
model_without_ddp, args.weight_decay)
optimizer = torch.optim.AdamW(param_groups, lr=args.lr, betas=(0.9, 0.95))
if misc.is_main_process():
print(optimizer)
loss_scaler = NativeScaler()
if args.resume and os.path.exists(args.resume):
print("we use resume student")
misc.load_model(args=args, model_without_ddp=model_without_ddp,
optimizer=optimizer, loss_scaler=loss_scaler)
if args.teacher_resume and os.path.exists(args.teacher_resume):
print("we use pretrain-teacher weight")
misc.load_teacher_model(
args=args, teacher_without_ddp=teacher_without_ddp)
else:
print("we use random initialized weight")
print(f"Start training for {args.epochs} epochs")
start_time = time.time()
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
data_loader_train.sampler.set_epoch(epoch)
train_stats = train_one_epoch(
model, teacher_without_ddp, data_loader_train,
optimizer, device, epoch, loss_scaler,
log_writer=log_writer,
args=args
)
log_stats = {**{f'train_{k}': v for k, v in train_stats.items()},
'epoch': epoch, }
if args.output_dir and misc.is_main_process():
# clean the information in log_writer
if log_writer is not None:
log_writer.flush()
with open(os.path.join(args.output_dir, "log.txt"), mode="a", encoding="utf-8") as f:
f.write(json.dumps(log_stats) + "\n")
if misc.is_main_process() and ((epoch+1) in args.stage_epochs):
misc.save_model(
args=args, model=model, model_without_ddp=model_without_ddp, optimizer=optimizer,
loss_scaler=loss_scaler, epoch=epoch)
misc.save_teacher_model(
args=args, model=teacher, model_without_ddp=teacher_without_ddp, epoch=epoch)
if (epoch+1) in args.stage_epochs and (epoch + 1) != args.epochs:
index = misc.find_stage_index(epoch + 1, args)
teacher = models_mae.__dict__[args.model](
depth=args.teacher_encoder_depth[index],
decoder_depth=0,
drop_path_rate=args.tdpr)
teacher.to(device)
teacher_without_ddp = teacher
for param in teacher.parameters():
param.requires_grad = False
msg = teacher_without_ddp.load_state_dict(
model.module.state_dict(), strict=False)
print(msg)
model.module.initialize_weights()
if misc.is_main_process() and (epoch+1) in args.stage_epochs:
print("Model = %s" % str(model_without_ddp))
print("teacher = %s" % str(teacher_without_ddp))
# save last model for future resume
if misc.is_main_process() and ((epoch+1) % 10 == 0):
misc.save_model(
args=args, model=model, model_without_ddp=model_without_ddp, optimizer=optimizer,
loss_scaler=loss_scaler, epoch=epoch, last=True)
misc.save_teacher_model(
args=args, model=teacher, model_without_ddp=teacher_without_ddp, last=1)
total_time = time.time() - start_time
total_time_str = str(datetime.timedelta(seconds=int(total_time)))
print('Training time {}'.format(total_time_str))
if __name__ == '__main__':
warnings.filterwarnings("ignore")
args = get_args_parser()
args = args.parse_args()
if args.output_dir:
Path(args.output_dir).mkdir(parents=True, exist_ok=True)
args.log_dir = args.output_dir
main(args)