-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathtrain.py
176 lines (152 loc) · 6.92 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
#!/usr/bin/python3
#coding=utf-8
from functools import partial
import sys
import datetime
import os
import time
import torch
import torch.nn.functional as F
from torch.utils.data import DataLoader
from tensorboardX import SummaryWriter
from data import dataset
import logging as logger
from lib.data_prefetcher import DataPrefetcher
import numpy as np
from train_processes import *
from tools import *
TAG = "scribblecod"
logger.basicConfig(level=logger.INFO, format='%(levelname)s %(asctime)s %(filename)s: %(lineno)d] %(message)s', datefmt='%Y-%m-%d %H:%M:%S', \
filename="train_%s.log"%(TAG), filemode="w")
import subprocess
GPU_ID = subprocess.getoutput('nvidia-smi --query-gpu=memory.free --format=csv,nounits,noheader | nl -v 0 | sort -nrk 2 | cut -f 1| head -n 1 | xargs')
os.environ['CUDA_VISIBLE_DEVICES'] = GPU_ID
""" set lr """
def get_triangle_lr(base_lr, max_lr, total_steps, cur, ratio=1., \
annealing_decay=1e-2, momentums=[0.95, 0.85]):
first = int(total_steps*ratio)
last = total_steps - first
min_lr = base_lr * annealing_decay
cycle = np.floor(1 + cur/total_steps)
x = np.abs(cur*2.0/total_steps - 2.0*cycle + 1)
if cur < first:
lr = base_lr + (max_lr - base_lr) * np.maximum(0., 1.0 - x)
else:
lr = ((base_lr - min_lr)*cur + min_lr*first - base_lr*total_steps)/(first - total_steps)
if isinstance(momentums, int):
momentum = momentums
else:
if cur < first:
momentum = momentums[0] + (momentums[1] - momentums[0]) * np.maximum(0., 1.-x)
else:
momentum = momentums[0]
return lr, momentum
def get_polylr(base_lr, last_epoch, num_steps, power):
return base_lr * (1.0 - min(last_epoch, num_steps-1) / num_steps) **power
def validate(model, val_loader):
model.train(False)
avg_mae = 0.0
cnt = 0
with torch.no_grad():
for image, mask, shape, name in val_loader:
image, mask = image.cuda().float(), mask.cuda().float()
out, _, _, _, _, _ = model(image)
out = F.interpolate(out, size=shape, mode='bilinear', align_corners=False)
pred = torch.sigmoid(out[0, 0])
pred = (pred - pred.min()) / (pred.max() - pred.min() + 1e-8)
avg_mae += torch.abs(pred - mask[0]).mean().item()
cnt += len(image)
model.train(True)
return (avg_mae / cnt)
def validate_multiloader(model, val_loader):
maes = []
for v in val_loader:
st = time.time()
mae = validate(model, v)
maes.append(mae)
print('Spent %.3fs, %s MAE: %s'%(time.time()-st, v.dataset.data_name, mae))
return sum(maes)/len(maes)
BASE_LR = 1e-5
MAX_LR = 1e-2
total_epoch = 150
EXP_NAME = '' # change it in main
root = '../CodDataset'
def train(Dataset, Network, cfg, train_loss, start_from = 0):
## dataset
data = Dataset.Data(cfg)
loader = DataLoader(data, batch_size=cfg.batch, shuffle=True, num_workers=8)
val_cfg = [Dataset.Config(datapath=f'{root}/test/{i}' , mode='test') for i in ['CHAMELEON', 'CAMO', 'COD10K']]
val_data = [Dataset.Data(v) for v in val_cfg]
val_loaders = [DataLoader(v, batch_size=1, shuffle=False, num_workers=4) for v in val_data]
min_mae = 1.0
best_epoch = 0
## network
net = Network(cfg)
# print('model has {} parameters in total'.format(sum(x.numel() for x in net.parameters())))
net.train(True)
net.cuda()
## parameter
base, head = [], []
for name, param in net.named_parameters():
if 'bkbone' in name:
base.append(param)
else:
head.append(param)
optimizer = torch.optim.SGD([{'params':base}, {'params':head}], lr=cfg.lr, momentum=cfg.momen, weight_decay=cfg.decay, nesterov=True)
## log
sw = SummaryWriter(cfg.savepath)
db_size = len(loader)
global_step = start_from * db_size
et = 0
# -------------------------- training ------------------------------------
# mae = validate_multiloader(net, val_loader)
# print(mae)
for epoch in range(start_from, cfg.epoch):
prefetcher = DataPrefetcher(loader)
batch_idx = -1
image, mask = prefetcher.next()
while image is not None:
st = time.time()
niter = epoch * db_size + batch_idx
lr, momentum = get_triangle_lr(BASE_LR, MAX_LR, cfg.epoch*db_size, niter, ratio=1.)
optimizer.param_groups[0]['lr'] = 0.1 * lr # for backbone
optimizer.param_groups[1]['lr'] = lr
optimizer.momentum = momentum
batch_idx += 1
global_step += 1
loss2, loss3, loss4, loss5, loss6 = train_loss(image, mask, net, dict(epoch=epoch+1, global_step=global_step, sw=sw, t_epo=cfg.epoch))
###### objective function ######
loss = loss2*1 + loss3*0.8 + loss4*0.6 + loss5*0.4 + loss6*0.2
optimizer.zero_grad()
loss.backward()
optimizer.step()
sw.add_scalar('lr', optimizer.param_groups[0]['lr'], global_step=global_step)
sw.add_scalar('loss', loss.item(), global_step=global_step)
image, mask = prefetcher.next()
ta = time.time() - st
et = 0.9*et + 0.1*ta if et>0 else ta
if batch_idx % 10 == 0:
msg = '%s| %s | eta:%s | step:%d/%d/%d | lr=%.6f | loss=%.6f | loss2=%.6f | loss3=%.6f | loss4=%.6f | loss5=%.6f' % (TAG, datetime.datetime.now(), datetime.timedelta(seconds = int((cfg.epoch*db_size-niter)*et)), global_step, epoch+1, cfg.epoch, optimizer.param_groups[0]['lr'], loss.item(), loss2.item(), loss3.item(), loss4.item(), loss5.item())
print(msg)
logger.info(msg)
if (epoch+1)%10==0:
mae = validate_multiloader(net, val_loaders)
print('VAL MAE:%s' % (mae))
sw.add_scalar('val', mae, global_step=global_step)
if mae < min_mae :
min_mae = mae
best_epoch = epoch + 1
if epoch > cfg.epoch//2:
torch.save(net.state_dict(), cfg.savepath + '/model-best.pth')
print('best epoch is:%d, MAE:%s' % (best_epoch, min_mae))
if epoch == cfg.epoch-2 or epoch == cfg.epoch-1 or (epoch+1) % 30 == 0:
torch.save(net.state_dict(), cfg.savepath + '/model-' + str(epoch + 1))
print('min val mae for {} is {}'.format(EXP_NAME, min_mae))
if __name__=='__main__':
cfg = [.15, 60, 16, 1]
w_ft, ft_st, topk,w_ftp = cfg
EXP_NAME = f'trained'
cfg = dataset.Config(datapath=f'{root}', savepath=f'./out/{EXP_NAME}/', mode='train', batch=16, lr=1e-3, momen=0.9, decay=5e-4, epoch=total_epoch, label_dir = 'Scribble')
from net import Net
tm = partial(train_loss, w_ft=w_ft, ft_st = ft_st, ft_fct=.5, ft_dct = dict(crtl_loss = False, w_ftp=w_ftp, norm=False, topk=topk, step_ratio=2), ft_head=False, mtrsf_prob=1, ops=[0,1,2], w_l2g=0.3, l_me=0.05, me_st=20, multi_sc=0)
train(dataset, Net, cfg, tm, start_from=0)