-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainer.py
256 lines (201 loc) · 10.2 KB
/
trainer.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
import os, sys
from tqdm import tqdm
import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torch.optim import SGD, Adam
from torch.optim.lr_scheduler import LambdaLR, CosineAnnealingLR
from models.sync_batchnorm.batchnorm import SynchronizedBatchNorm2d
from models.sync_batchnorm.replicate import patch_replication_callback
from models import SwiftNet, DeepLab
from dataloaders import CityscapesBase
from utils import Saver, Evaluator, TensorboardSummary
class Trainer(object):
def __init__(self, args):
self.args = args
self.logger = args.logger
self.device = torch.device(args.gpu_ids[0] if args.cuda else 'cpu')
# Define Dataloader
self.train_loader, self.val_loader, num_classes = self.get_data_loaders()
# Define Network
model, train_params = self.get_model_and_params(num_classes)
self.model = model.to(self.device)
self.criterion = nn.CrossEntropyLoss(ignore_index=255, reduction='mean')
self.evaluator = Evaluator(num_classes)
# Using DataParallel
if args.cuda:
self.model = nn.DataParallel(model, device_ids=self.args.gpu_ids)
patch_replication_callback(self.model)
if args.eval_only:
self.init_eval_mode()
else:
self.init_train_mode(train_params)
def init_train_mode(self, train_params):
args = self.args
# Define Saver and Tensorboard Summary
self.saver = Saver(args)
self.saver.save_experiment_config()
self.writer = TensorboardSummary(self.saver.experiment_dir, self.val_loader.dataset)
self.logger.info('\nExperiment dir : %s', self.saver.experiment_dir)
# Define Optimizer and Scheduler
if args.optim == 'sgd':
self.optimizer = SGD(train_params, weight_decay=args.weight_decay,
momentum=args.momentum, nesterov=args.nesterov)
elif args.optim == 'adam':
self.optimizer = Adam(train_params, weight_decay=args.weight_decay)
if args.scheduler == 'poly':
update_func = lambda epoch: (1 - 1.0 * epoch / args.epochs) ** args.poly_power
self.scheduler = LambdaLR(self.optimizer, update_func)
elif args.scheduler == 'cos':
self.scheduler = CosineAnnealingLR(self.optimizer, args.epochs, eta_min=args.cos_eta_min)
# Resuming checkpoint
self.best_pred = 0.0
self.args.start_epoch = 0
if args.resume is not None:
if not os.path.isfile(args.resume):
raise RuntimeError("No checkpoint found at '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch'] + 1
if args.cuda:
self.model.module.load_state_dict(checkpoint['state_dict'])
else:
self.model.load_state_dict(checkpoint['state_dict'])
if not args.ft:
if 'scheduler' in checkpoint.keys():
self.scheduler.load_state_dict(checkpoint['scheduler'])
else:
print("ALERT: No scheduler found in checkpoint")
for _ in range(args.start_epoch): self.scheduler.step()
self.optimizer.load_state_dict(checkpoint['optimizer'])
self.best_pred = checkpoint['best_pred']
self.logger.info("Loaded checkpoint '%s' (epoch %s)", args.resume, checkpoint['epoch'])
# Clear start epoch if fine-tuning
if args.ft:
args.start_epoch = 0
self.logger.info('Using %s architecture' % args.archi)
self.logger.info('Using %s backbone' % args.backbone)
self.logger.info('Using %s optimizer' % args.optim)
self.logger.info('Using %s scheduler' % args.scheduler)
self.logger.info('Using %s images / batch [train]' % args.batch_size)
self.logger.info('Using %s images / batch [val]' % args.val_batch_size)
def init_eval_mode(self):
args = self.args
# Resuming checkpoint
if args.resume is not None:
if not os.path.isfile(args.resume):
raise RuntimeError("No checkpoint found at '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
if args.cuda:
self.model.module.load_state_dict(checkpoint['state_dict'])
else:
self.model.load_state_dict(checkpoint['state_dict'])
self.logger.info("Loaded checkpoint '%s' (epoch %s)", args.resume, checkpoint['epoch'])
self.logger.info('Using %s architecture' % args.archi)
self.logger.info('Using %s backbone' % args.backbone)
self.logger.info('Using %s images / batch [val]' % args.val_batch_size)
def get_data_loaders(self):
kwargs = {'num_workers': self.args.workers, 'pin_memory': True}
train_set = CityscapesBase(**self.args.db, split='train')
val_set = CityscapesBase(**self.args.db, split='val')
train_loader = DataLoader(train_set, batch_size=self.args.batch_size, shuffle=True, **kwargs)
val_loader = DataLoader(val_set, batch_size=self.args.val_batch_size, shuffle=False, **kwargs)
return train_loader, val_loader, train_set.num_classes
def get_model_and_params(self, num_classes):
args = self.args
norm_layer = SynchronizedBatchNorm2d if args.sync_bn else nn.BatchNorm2d
if args.archi == 'deeplab':
model = DeepLab(num_classes=num_classes,
backbone=args.backbone,
output_stride=args.out_stride,
norm_layer=norm_layer,
freeze_bn=args.freeze_bn,
image_lvl_feat=not args.no_image_lvl_feat)
train_params = [{'params': model.fine_tune_params(), 'lr': args.lr / 10, 'weight_decay': args.weight_decay / 10},
{'params': model.random_init_params(), 'lr': args.lr, 'weight_decay': args.weight_decay}]
elif args.archi == 'swiftnet':
model = SwiftNet(backbone=args.backbone, num_classes=num_classes)
train_params = [{'params': model.random_init_params(), 'lr': args.lr, 'weight_decay': args.weight_decay},
{'params': model.fine_tune_params(), 'lr': args.lr / 4, 'weight_decay': args.weight_decay / 4}]
else:
raise NotImplementedError("Architecture {} is not implemented".format(args.archi))
return model, train_params
def run(self):
if self.args.eval_only:
self.validate()
else:
self.train()
def train(self):
self.logger.info('\n' + '-' * 20)
self.logger.info('Starting Training')
self.logger.info('Starting Epoch: %d' % self.args.start_epoch)
self.logger.info('Total Epoches: %d\n' % self.args.epochs)
for epoch in range(self.args.start_epoch, self.args.epochs):
is_best = False
is_last = (epoch + 1) == self.args.epochs
self.train_epoch(epoch)
if (epoch + 1) % self.args.eval_interval == 0 or is_last:
metrics = self.validate(epoch)
self.writer.write_metrics(metrics, epoch)
is_best = metrics['mIoU'] > self.best_pred
self.best_pred = max(metrics['mIoU'], self.best_pred)
if is_best or self.args.save_every_epoch or is_last:
self.save(epoch, is_best=is_best)
self.writer.close()
def train_epoch(self, epoch):
torch.set_grad_enabled(True)
self.model.train()
train_loss = 0
tbar = tqdm(self.train_loader, file=sys.stdout)
num_img_tr = len(self.train_loader)
self.train_loader.dataset.seed = epoch
for i, sample in enumerate(tbar):
loss, image, target, output = self.process_sample(sample, index=i)
train_loss += loss
if i % 50 == 0:
tbar.set_description('Train loss: %.3f' % (train_loss / (i + 1)))
self.writer.add_scalar('train/total_loss_iter', loss, i + num_img_tr * epoch)
# Show inference results each half of epoch
if i % (num_img_tr // 2) == 0:
global_step = i + num_img_tr * epoch
self.writer.visualize_image(image, target, output, global_step)
self.scheduler.step()
self.writer.add_scalar('train/total_loss_epoch', train_loss, epoch)
self.logger.info('[End of Epoch %d]' % epoch)
self.logger.info('Loss: %.3f' % train_loss)
def validate(self, epoch=0):
torch.set_grad_enabled(False)
self.evaluator.reset()
self.model.eval()
test_loss = 0.0
tbar = tqdm(self.val_loader, desc='\r', file=sys.stdout)
for i, sample in enumerate(tbar):
loss, image, target, output = self.process_sample(sample, index=i, train=False)
test_loss += loss
tbar.set_description('Test loss: %.3f' % (test_loss / (i + 1)))
pred = output.argmax(dim=1).cpu().numpy()
target = target.cpu().numpy()
self.evaluator.add_batch(target, pred)
metrics = self.evaluator.get_all_metrics()
metrics['total_loss_epoch'] = test_loss
self.logger.info('Validation: [Epoch: %d]' % epoch)
self.logger.info("Acc:{Acc}, Acc_class:{Acc_class}, mIoU:{mIoU}, fwIoU: {fwIoU}".format(**metrics))
self.logger.info('Loss: %.3f' % test_loss)
return metrics
def process_sample(self, sample, index, *, train=True):
image = sample['image_0'].to(self.device)
target = sample['label'].to(self.device)
output = self.model(image)
loss = self.criterion(output, target)
if train:
loss.backward()
self.optimizer.step()
self.optimizer.zero_grad()
return loss.item(), image, target, output
def save(self, epoch, is_best=False):
self.saver.save_checkpoint({
'epoch': epoch,
'state_dict': self.model.module.state_dict(),
'optimizer': self.optimizer.state_dict(),
'scheduler': self.scheduler.state_dict(),
'best_pred': self.best_pred,
}, is_best=is_best)