-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathtrain.py
167 lines (158 loc) · 7.68 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
import torch
from torch import nn
from torch.autograd import Variable
import time
import os
from validation import val_epoch
from model import generate_model
from opts import parse_opts
from utils import AverageMeter, accuracy, adjust_learning_rate, save_checkpoint
from dataset import Video
from spatial_transforms import (Compose, Normalize, Scale,
CenterCrop, CornerCrop, MultiScaleCornerCrop,
MultiScaleRandomCrop, RandomHorizontalFlip, ToTensor)
from temporal_transforms import TemporalRandomCrop, LoopPadding
import numpy as np
from torch.optim import lr_scheduler
from mean import get_mean
from collections import defaultdict
best_prec1 = 0
def train_epoch(epoch, data_loader, model, criterion, optimizer, opt):
print('train at epoch {} with lr {}'.format(epoch, optimizer.param_groups[-1]['lr']))
model.train()
batch_time = AverageMeter()
data_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
end_time = time.time()
for i, (inputs, targets, _) in enumerate(data_loader):
data_time.update(time.time() - end_time)
if not opt.no_cuda:
targets = targets.cuda(async=True)
inputs = Variable(inputs)
targets_var = Variable(targets)
outputs = model(inputs)
loss = criterion(outputs, targets_var)
prec1, prec5 = accuracy(outputs.data, targets, topk=(1, 5))
losses.update(loss.data[0], inputs.size(0))
top1.update(prec1[0], inputs.size(0))
top5.update(prec5[0], inputs.size(0))
optimizer.zero_grad()
loss.backward()
optimizer.step()
batch_time.update(time.time() - end_time)
end_time = time.time()
if i % opt.print_freq == 0:
print('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
epoch, i + 1, len(data_loader), batch_time=batch_time,
data_time=data_time, loss=losses, top1=top1, top5=top5))
def validate(val_loader, model, criterion):
batch_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
model.eval()
end = time.time()
for i, (input, target, _) in enumerate(val_loader):
target = target.cuda(async=True)
input_var = torch.autograd.Variable(input, volatile=True)
target_var = torch.autograd.Variable(target, volatile=True)
output = model(input_var)
loss = criterion(output, target_var)
prec1, prec5 = accuracy(output.data, target, topk=(1,5))
losses.update(loss.data[0], input.size(0))
top1.update(prec1[0], input.size(0))
top5.update(prec5[0], input.size(0))
batch_time.update(time.time() - end)
end = time.time()
if i % opt.print_freq == 0:
print(('Test: [{0}/{1}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
i, len(val_loader), batch_time=batch_time, loss=losses,
top1=top1, top5=top5)))
print(('Testing Results: Prec@1 {top1.avg:.3f} Prec@5 {top5.avg:.3f} Loss {loss.avg:.5f}'
.format(top1=top1, top5=top5, loss=losses)))
return losses.avg, top1.avg
if __name__ == '__main__':
opt = parse_opts()
opt.mean = get_mean(1)
opt.arch = '{}-{}'.format(opt.model_name, opt.model_depth)
opt.sample_duration = 16
opt.scales = [opt.initial_scale]
for i in range(1, opt.n_scales):
opt.scales.append(opt.scales[-1] * opt.scale_step)
print('#####', opt.scales)
print(opt.mean)
spatial_transform = Compose([MultiScaleCornerCrop(opt.scales, opt.sample_size),
RandomHorizontalFlip(),
ToTensor(1),
Normalize(opt.mean, [1, 1, 1])])
temporal_transform = TemporalRandomCrop(opt.sample_duration)
train_data = Video(opt.train_list, spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
sample_duration=opt.sample_duration, n_samples_for_each_video=1)
train_loader = torch.utils.data.DataLoader(train_data, batch_size=opt.batch_size, shuffle=True, num_workers=opt.n_threads, pin_memory=True)
val_spatial_transform = Compose([Scale(opt.sample_size),
CenterCrop(opt.sample_size),
ToTensor(1),
Normalize(opt.mean, [1, 1, 1])])
val_temporal_transform = LoopPadding(opt.sample_duration)
val_data = Video(opt.val_list, spatial_transform=val_spatial_transform,
temporal_transform=val_temporal_transform,
sample_duration=opt.sample_duration, n_samples_for_each_video=opt.n_val_samples)
val_loader = torch.utils.data.DataLoader(val_data, batch_size=opt.batch_size, shuffle=False, num_workers=opt.n_threads, pin_memory=True)
model, policies = generate_model(opt)
model = nn.DataParallel(model, device_ids=opt.gpus).cuda()
if opt.finetune:
if os.path.isfile(opt.finetune):
print('finetuning from model {}'.format(opt.finetune))
model_data = torch.load(opt.finetune)
own_state = model.state_dict()
for k, v in model_data['state_dict'].items():
if 'fc' in k:
continue
print(k)
if isinstance(v, torch.nn.parameter.Parameter):
v = v.data
assert v.dim() == own_state[k].dim(), '{} {} vs {}'.format(k, v.dim(), own_state[k].dim())
own_state[k].copy_(v)
else:
assert False, ("=> no checkpoint found at '{}'".format(opt.finetune))
if opt.resume:
if os.path.isfile(opt.resume):
print('loading model {}'.format(opt.resume))
model_data = torch.load(opt.resume)
opt.start_epoch = model_data['epoch']
best_prec1 = model_data['best_prec1']
model.load_state_dict(model_data['state_dict'])
else:
assert False,("=> no checkpoint found at '{}'".format(opt.resume))
criterion = torch.nn.CrossEntropyLoss().cuda()
optimizer = torch.optim.SGD(policies, opt.lr, momentum=opt.momentum, dampening=opt.dampening,
weight_decay=opt.weight_decay, nesterov=opt.nesterov)
scheduler = lr_scheduler.ReduceLROnPlateau(optimizer, 'min', patience=opt.lr_patience)
for epoch in range(opt.start_epoch, opt.epochs):
#adjust_learning_rate(optimizer, epoch, opt.lr_steps, opt)
train_epoch(epoch, train_loader, model, criterion, optimizer, opt)
if (epoch + 1) % opt.eval_freq == 0 or epoch == opt.epochs - 1:
loss, prec1 = validate(val_loader, model, criterion)
is_best = prec1 > best_prec1
best_prec1 = max(prec1, best_prec1)
save_checkpoint({
'epoch': epoch + 1,
'arch': opt.arch,
'state_dict': model.state_dict(),
'optimizer' : optimizer.state_dict(),
'best_prec1': best_prec1,
}, is_best, opt.snapshot_pref)
print('best_prec1: ', best_prec1)
scheduler.step(loss)