-
Notifications
You must be signed in to change notification settings - Fork 9
/
train.py
279 lines (227 loc) · 10.4 KB
/
train.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
from __future__ import division
import os
import argparse
import time
from copy import deepcopy
import torch
import torch.distributed as dist
from torch.nn.parallel import DistributedDataParallel as DDP
from utils import distributed_utils
from utils.com_flops_params import FLOPs_and_Params
from utils.misc import CollateFunc, get_total_grad_norm, vis_data
from utils.misc import build_dataset, build_dataloader
from utils.solver.optimizer import build_optimizer
from utils.solver.warmup_schedule import build_warmup
from models import build_model
from config import build_config
def parse_args():
parser = argparse.ArgumentParser(description='YOLOF Detection')
# basic
parser.add_argument('--cuda', action='store_true', default=False,
help='use cuda.')
parser.add_argument('-bs', '--batch_size', default=16, type=int,
help='Batch size on single GPU for training')
parser.add_argument('--schedule', type=str, default='1x',
help='training schedule: 1x, 2x, 3x, ...')
parser.add_argument('--num_workers', default=4, type=int,
help='Number of workers used in dataloading')
parser.add_argument('--eval_epoch', default=1, type=int,
help='interval between evaluations')
parser.add_argument('--grad_clip_norm', default=-1., type=float,
help='grad clip.')
parser.add_argument('--tfboard', action='store_true', default=False,
help='use tensorboard')
parser.add_argument('--save_folder', default='weights/', type=str,
help='path to save weight')
parser.add_argument('--vis', dest="vis", action="store_true", default=False,
help="visualize input data.")
# model
parser.add_argument('-v', '--version', default='yolof-r50',
help='build object detector')
parser.add_argument('--topk', default=1000, type=int,
help='NMS threshold')
parser.add_argument('-p', '--coco_pretrained', default=None, type=str,
help='coco pretrained weight')
# dataset
parser.add_argument('--root', default='/mnt/share/ssd2/dataset',
help='data root')
parser.add_argument('-d', '--dataset', default='coco',
help='coco, voc, widerface, crowdhuman')
# train trick
parser.add_argument('--no_warmup', action='store_true', default=False,
help='do not use warmup')
# DDP train
parser.add_argument('-dist', '--distributed', action='store_true', default=False,
help='distributed training')
parser.add_argument('--dist_url', default='env://',
help='url used to set up distributed training')
parser.add_argument('--world_size', default=1, type=int,
help='number of distributed processes')
parser.add_argument('--sybn', action='store_true', default=False,
help='use sybn.')
return parser.parse_args()
def train():
args = parse_args()
print("Setting Arguments.. : ", args)
print("----------------------------------------------------------")
# dist
if args.distributed:
distributed_utils.init_distributed_mode(args)
print("git:\n {}\n".format(distributed_utils.get_sha()))
# path to save model
path_to_save = os.path.join(args.save_folder, args.dataset, args.version)
os.makedirs(path_to_save, exist_ok=True)
# cuda
if args.cuda:
print('use cuda')
device = torch.device("cuda")
else:
device = torch.device("cpu")
# YOLOF Config
cfg = build_config(args)
print('==============================')
print('Model Configuration: \n', cfg)
# dataset and evaluator
dataset, evaluator, num_classes = build_dataset(cfg, args, device)
# dataloader
dataloader = build_dataloader(args, dataset, args.batch_size, CollateFunc())
# build model & criterion
model, criterion = build_model(args=args, cfg=cfg, device=device, num_classes=num_classes, trainable=True)
model = model.to(device).train()
# DDP
model_without_ddp = model
if args.distributed:
model = DDP(model, device_ids=[args.gpu])
model_without_ddp = model.module
# optimizer
base_lr = cfg['base_lr'] * args.batch_size * distributed_utils.get_world_size()
backbone_lr = base_lr * cfg['bk_lr_ratio']
optimizer = build_optimizer(cfg, model_without_ddp, base_lr, backbone_lr)
# lr scheduler
lr_epoch = cfg['epoch'][args.schedule]['lr_epoch']
lr_scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer=optimizer, milestones=lr_epoch)
# warmup scheduler
warmup_scheduler = build_warmup(cfg, base_lr)
# training configuration
max_epoch = cfg['epoch'][args.schedule]['max_epoch']
epoch_size = len(dataloader)
best_map = -1.
warmup = not args.no_warmup
# compute FLOPs and Params
if distributed_utils.is_main_process:
model_copy = deepcopy(model_without_ddp)
FLOPs_and_Params(model=model_copy,
min_size=cfg['test_min_size'],
max_size=cfg['test_max_size'],
device=device)
del model_copy
if args.distributed:
# wait for all processes to synchronize
dist.barrier()
t0 = time.time()
# start training loop
for epoch in range(max_epoch):
if args.distributed:
dataloader.batch_sampler.sampler.set_epoch(epoch)
# train one epoch
for iter_i, (images, targets, masks) in enumerate(dataloader):
ni = iter_i + epoch * epoch_size
# warmup
if ni < cfg['wp_iter'] and warmup:
warmup_scheduler.warmup(ni, optimizer)
elif ni == cfg['wp_iter'] and warmup:
# warmup is over
print('Warmup is over')
warmup = False
warmup_scheduler.set_lr(optimizer, base_lr, base_lr)
# to device
images = images.to(device)
masks = masks.to(device)
targets = [{k: v.to(device) for k, v in t.items()} for t in targets]
# visualize input data
if args.vis:
vis_data(images, targets, masks)
continue
# inference
outputs = model(images, mask=masks)
# compute loss
loss_dict = criterion(outputs, targets)
losses = loss_dict['total_loss']
loss_dict_reduced = distributed_utils.reduce_dict(loss_dict)
# check loss
if torch.isnan(losses):
print('loss is NAN !!')
continue
# Backward and Optimize
losses.backward()
if args.grad_clip_norm > 0.:
total_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip_norm)
else:
total_norm = get_total_grad_norm(model.parameters())
optimizer.step()
optimizer.zero_grad()
# display
if distributed_utils.is_main_process() and iter_i % 10 == 0:
t1 = time.time()
cur_lr = [param_group['lr'] for param_group in optimizer.param_groups]
cur_lr_dict = {'lr': cur_lr[0], 'lr_bk': cur_lr[1]}
# basic infor
log = '[Epoch: {}/{}]'.format(epoch+1, max_epoch)
log += '[Iter: {}/{}]'.format(iter_i, epoch_size)
log += '[lr: {:.6f}][lr_bk: {:.6f}]'.format(cur_lr_dict['lr'], cur_lr_dict['lr_bk'])
# loss infor
for k in loss_dict_reduced.keys():
log += '[{}: {:.2f}]'.format(k, loss_dict_reduced[k])
# other infor
log += '[time: {:.2f}]'.format(t1 - t0)
log += '[gnorm: {:.2f}]'.format(total_norm)
log += '[size: [{}, {}]]'.format(cfg['train_min_size'], cfg['train_max_size'])
# print log infor
print(log, flush=True)
t0 = time.time()
lr_scheduler.step()
# evaluation
if epoch % args.eval_epoch == 0 or (epoch + 1) == max_epoch:
# check evaluator
if distributed_utils.is_main_process():
if evaluator is None:
print('No evaluator ... save model and go on training.')
print('Saving state, epoch: {}'.format(epoch + 1))
weight_name = '{}_epoch_{}.pth'.format(args.version, epoch + 1)
checkpoint_path = os.path.join(path_to_save, weight_name)
torch.save({'model': model_without_ddp.state_dict(),
'epoch': epoch,
'args': args},
checkpoint_path)
else:
print('eval ...')
model_eval = model_without_ddp
# set eval mode
model_eval.trainable = False
model_eval.eval()
# evaluate
evaluator.evaluate(model_eval)
cur_map = evaluator.map
if cur_map > best_map:
# update best-map
best_map = cur_map
# save model
print('Saving state, epoch:', epoch + 1)
weight_name = '{}_epoch_{}_{:.2f}.pth'.format(args.version, epoch + 1, best_map*100)
checkpoint_path = os.path.join(path_to_save, weight_name)
torch.save({'model': model_without_ddp.state_dict(),
'epoch': epoch,
'args': args},
checkpoint_path)
# set train mode.
model_eval.trainable = True
model_eval.train()
if args.distributed:
# wait for all processes to synchronize
dist.barrier()
# close mosaic augmentation
if cfg['mosaic'] and max_epoch - epoch == 5:
print('close Mosaic Augmentation ...')
dataloader.dataset.mosaic = False
if __name__ == '__main__':
train()