-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtrainer_ACDC.py
187 lines (160 loc) · 8.12 KB
/
trainer_ACDC.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
#!/usr/bin/env python
# -*- coding:utf-8 -*-
import sys
import logging
import torch
import torch.nn as nn
import torch.optim as optim
from torch.nn.modules.loss import CrossEntropyLoss
import torchvision
# import matplotlib.pyplot as plt
from utils.utils import DiceLoss
from torch.utils.data import DataLoader
from datasets.dataset_ACDC import ACDCdataset, RandomGenerator
from tqdm import tqdm
import os
from torchvision import transforms
from utils.test_ACDC import inference
from medpy.metric import dc,hd95
def create_lr_scheduler(optimizer,
num_step: int,
epochs: int,
warmup=True,
warmup_epochs=15,
warmup_factor=1e-3):
assert num_step > 0 and epochs > 0
if warmup is False:
warmup_epochs = 0
def f(x):
"""
根据step数返回一个学习率倍率因子，
注意在训练开始之前，pytorch会提前调用一次lr_scheduler.step()方法
"""
warmup_epochs = int(epochs / 20)
if warmup is True and x <= (warmup_epochs * num_step):
alpha = float(x) / (warmup_epochs * num_step)
# warmup过程中lr倍率因子从warmup_factor -> 1
return warmup_factor * (1 - alpha) + alpha
else:
# warmup后lr倍率因子从1 -> 0
# 参考deeplab_v2: Learning rate policy
return (1 - (x - warmup_epochs * num_step) / ((epochs - warmup_epochs) * num_step)) ** 0.9
return torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=f)
def trainer_acdc(args, model, snapshot_path):
# if args.usecheckpoint:
# model.load_state_dict(torch.load(args.checkpoint))
train_dataset = ACDCdataset(args.root_path, args.list_dir, split="train", transform=
transforms.Compose(
[RandomGenerator(output_size=[args.img_size, args.img_size])]))
Train_loader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True)
db_val=ACDCdataset(base_dir=args.root_path, list_dir=args.list_dir, split="valid")
valloader=DataLoader(db_val, batch_size=1, shuffle=False)
db_test =ACDCdataset(base_dir=args.volume_path,list_dir=args.list_dir, split="test")
testloader = DataLoader(db_test, batch_size=1, shuffle=False)
if args.n_gpu > 1:
model = nn.DataParallel(model)
model = model.cuda()
model.train()
ce_loss = CrossEntropyLoss()
dice_loss = DiceLoss(args.num_classes)
save_interval = args.n_skip # int(max_epoch/6)
iter_num = 0
Loss = []
Test_Accuracy = []
Best_dcs = 0.8
logging.basicConfig(filename=snapshot_path + "/log.txt", level=logging.INFO, format='%(asctime)s %(levelname)s %(message)s')
logging.getLogger().addHandler(logging.StreamHandler(sys.stdout))
logging.info(str(args))
max_iterations = args.max_epochs * len(Train_loader)
base_lr = args.base_lr
base_weight = args.base_weight
optimizer = optim.AdamW(model.parameters(), lr=base_lr, weight_decay=base_weight)
# optimizer = optim.Adam(model.parameters(), lr=base_lr, weight_decay=0.0001)
# optimizer = optim.SGD(model.parameters(), lr=base_lr, momentum=0.9, weight_decay=0.0001)
# 创建学习率更新策略，这里是每个step更新一次(不是每个epoch)
lr_scheduler = create_lr_scheduler(optimizer, len(Train_loader), args.max_epochs, warmup=True)
for epoch in range(args.max_epochs):
model.train()
train_loss = 0
with tqdm(desc='Epoch %d - train' % (epoch),
unit='it', total=len(Train_loader)) as pbar:
for i_batch, sampled_batch in enumerate(Train_loader):
image_batch, label_batch = sampled_batch["image"], sampled_batch["label"]
image_batch, label_batch = image_batch.type(torch.FloatTensor), label_batch.type(torch.FloatTensor)
image_batch, label_batch = image_batch.cuda(), label_batch.cuda()
outputs = model(image_batch)
loss_ce = ce_loss(outputs, label_batch[:].long())
loss_dice = dice_loss(outputs, label_batch[:], softmax=True)
loss = loss_dice * 0.5+ loss_ce * 0.5
optimizer.zero_grad()
loss.backward()
optimizer.step()
lr_scheduler.step()
lr_ = optimizer.param_groups[0]["lr"]
# lr_ = base_lr * (1.0 - iter_num / max_iterations) ** 0.9
# for param_group in optimizer.param_groups:
# param_group['lr'] = lr_
iter_num = iter_num + 1
train_loss += loss.item()
# logging.info('iteration %d : loss : %f, loss_ce: %f' % (iter_num, loss.item(), loss_ce.item()))
pbar.set_postfix(loss=train_loss / (i_batch + 1), lr=lr_)
pbar.update()
# Loss.append(train_loss/len(train_dataset))
# loss visualization
# fig1, ax1 = plt.subplots(figsize=(11, 8))
# ax1.plot(range(epoch + 1), Loss)
# ax1.set_title("Average trainset loss vs epochs")
# ax1.set_xlabel("Epoch")
# ax1.set_ylabel("Current loss")
# plt.savefig('loss_vs_epochs_gauss.png')
# plt.clf()
# plt.close()
# ---------- Validation ----------
if epoch > 50 and (epoch + 1) % save_interval == 0:
dc_sum = 0
model.eval()
for i, val_sampled_batch in enumerate(valloader):
val_image_batch, val_label_batch = val_sampled_batch["image"], val_sampled_batch["label"]
val_image_batch, val_label_batch = val_image_batch.type(torch.FloatTensor), val_label_batch.type(
torch.FloatTensor)
val_image_batch, val_label_batch = val_image_batch.cuda().unsqueeze(
1), val_label_batch.cuda().unsqueeze(1)
val_outputs = model(val_image_batch)
val_outputs = torch.argmax(torch.softmax(val_outputs, dim=1), dim=1).squeeze(0)
dc_sum += dc(val_outputs.cpu().data.numpy(), val_label_batch[:].cpu().data.numpy())
avg_dcs = dc_sum / len(valloader)
logging.info("Validation ===>avg_dsc: %f" % avg_dcs)
if avg_dcs > Best_dcs:
save_mode_path = os.path.join(snapshot_path, 'epoch={}_avg_dcs={}.pth'.format(epoch, avg_dcs))
torch.save(model.state_dict(), save_mode_path)
logging.info("save model to {}".format(save_mode_path))
Best_dcs = avg_dcs
# ---------- Test ----------
avg_dcs, avg_hd = inference(args, model, testloader, args.test_save_dir) #args.test_save_dir
Test_Accuracy.append(avg_dcs)
elif avg_dcs > 0.83:
# ---------- Test ----------
avg_dcs, avg_hd = inference(args, model, testloader, None)
Test_Accuracy.append(avg_dcs)
save_mode_path = os.path.join(snapshot_path, 'test_epoch={}_avg_dcs={}.pth'.format(epoch, avg_dcs))
torch.save(model.state_dict(), save_mode_path)
logging.info("save model to {}".format(save_mode_path))
# val visualization
# fig2, ax2 = plt.subplots(figsize=(11, 8))
# ax2.plot(range(int((epoch + 1) // save_interval)), Test_Accuracy)
# ax2.set_title("Average val dataset dice score vs epochs")
# ax2.set_xlabel("Epoch")
# ax2.set_ylabel("Current dice score")
# plt.savefig('val_dsc_vs_epochs_gauss.png')
# plt.clf()
# plt.close()
if epoch >= args.max_epochs - 1:
save_mode_path = os.path.join(snapshot_path, 'epoch={}_lr={}.pth'.format(epoch, lr_))
torch.save(model.state_dict(), save_mode_path)
logging.info("save model to {}".format(save_mode_path))
# ---------- Test ----------
avg_dcs, avg_hd = inference(args, model, testloader, None)
Test_Accuracy.append(avg_dcs)
print(max(Test_Accuracy))
break
return "Training Finished!"