-
Notifications
You must be signed in to change notification settings - Fork 1
/
go_on_train.py
120 lines (100 loc) · 4.07 KB
/
go_on_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
import torch
from torch import nn
from torch.utils.data import DataLoader
from torch import optim
from matplotlib import pyplot as plt
# import numpy as np
from datetime import datetime
# from pathlib import Path
# from torch.cuda.amp import GradScaler, autocast
import os
import gc
from data_handle import MyData
from LiteSeg import liteseg
from utils import get_parse
gc.collect()
torch.cuda.empty_cache()
args = get_parse()
cur_model_name = args.weight.split('/')[-1].split('.')[0]
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
train_dataset = MyData(args.train_label, args.train_data, img_size=(640, 640))
train_dataloader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True, num_workers=args.workers)
# scaler = GradScaler()
go_on_weight = args.checkpoint_path + args.go_on_param
save_checkpoint_path = args.checkpoint_path + cur_model_name + '/'
checkpoints = torch.load(go_on_weight)
model = liteseg.LiteSeg(num_class=1, backbone_network='mobilenet', pretrain_weight=None, is_train=False)
model.load_state_dict(checkpoints['model_state_dict'])
# print(model)
loss_func = nn.BCELoss()
optimizer = optim.Adam(model.parameters())
optimizer.load_state_dict(checkpoints['optimizer_state_dict'])
# loss_func = nn.BCEWithLogitsLoss()
# optimizer = optim.AdamW(model.parameters())
def draw_loss(loss_list, epochs):
x = [_ for _ in range(len(loss_list))]
y = loss_list
plt.figure()
plt.xlabel('epoch')
plt.ylabel('loss')
# plt.scatter(x, y)
plt.legend(loc='best', labels=['loss'])
plt.title('unet train loss curve')
plt.plot(x, y, label='unet train loss curve')
plt.savefig(os.path.join(args.train_loss_curve_save_path + cur_model_name, '_epoch_' + str(epochs) + '.png'))
print(str(epochs) + ' loss save ok')
plt.close()
def mkdir_path():
# f = Path(__file__).resolve()
# root = f.parents[0]
if not os.path.exists(save_checkpoint_path):
os.makedirs(save_checkpoint_path)
if not os.path.exists(args.train_loss_curve_save_path + cur_model_name):
os.makedirs(args.train_loss_curve_save_path + cur_model_name)
def train():
model.train()
best_loss = 0.0001
train_loss_list = []
for i in range(args.go_on_epoch, 1 + args.epochs):
epoch_loss = 0
step = 0
for data, label in train_dataloader:
optimizer.zero_grad()
data, label = data.to(device), label.to(device)
# with autocast():
output = model(data)
loss = loss_func(output, label)
# scaler.scale(loss).backward() # 将张量乘以比例因子,反向传播
# scaler.step(optimizer) # 将优化器的梯度张量除以比例因子。
# scaler.update() # 更新比例因子
loss.backward()
optimizer.step()
epoch_loss += loss.item()
step += 1
if step % 5 == 0:
print('[%s]: Epoch %d ======> global step %d/%d ===========> train_loss: %.6f ' % (
datetime.now(), i, step, len(train_dataloader), loss.item()))
if best_loss > loss.item():
torch.save(model.state_dict(), args.weight)
best_loss = loss.item()
print('cur best loss: ', best_loss, '\tsave ok')
# 每20次epoch存一次断点
if i % 20 == 0:
checkpoint = {
'epochs': i,
'model_state_dict': model.state_dict(),
'optimizer_state_dict': optimizer.state_dict()
}
torch.save(checkpoint, save_checkpoint_path + '_ep' + str(i) + '.pth')
print('ep_', i, '_pth : =================== checkpoint save ok')
avg_loss = epoch_loss / len(train_dataloader)
print("Epoch: %d ===========> Avg loss: %.4f " % (i, avg_loss))
train_loss_list.append(avg_loss)
if i % 5 == 0:
draw_loss(train_loss_list, i)
if i == args.epochs:
torch.save(model.state_dict(), args.weight_last)
print('last model save ok!')
if __name__ == '__main__':
mkdir_path()
train()