-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
186 lines (140 loc) · 6.81 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
177
178
179
180
181
182
183
184
185
186
import monai
from dataloader.get_dataloaders import get_dataloaders
import torch
from monai.losses.dice import DiceLoss
import wandb
from tqdm import tqdm
import torch.nn as nn
from Metrics.calculate_metrics import calculate_metrics
from operator import add
import pandas as pd
import os
def training_model(train_dataloader,val_dataloader, dropout, epochs, learning_rate):
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = monai.networks.nets.UNet(
spatial_dims=3, # 2 or 3 for a 2D or 3D network
in_channels=1, # number of input channels
out_channels=3, # number of output channels
channels=[8, 16, 32], # channel counts for layers
strides=[2, 2], # strides for mid layers
dropout = dropout
).to(device)
num_epochs = epochs
loss_function = DiceLoss() # using my own dice loss
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
checkpoint_path = 'model.pth'
start_epoch = 1
if os.path.exists(checkpoint_path):
checkpoint = torch.load(checkpoint_path)
model.load_state_dict(checkpoint['model_state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
start_epoch = checkpoint['epoch'] + 1
print(f"Resuming training from epoch {start_epoch}")
for epoch in range(start_epoch,num_epochs+1):
overall_train_loss_per_epoch = []
overall_val_loss_per_epoch = []
overall_train_jaccard_per_epoch = []
overall_val_jaccard_per_epoch = []
overall_train_acc_per_epoch = []
overall_val_acc_per_epoch = []
epoch_train_loss = 0.0
epoch_val_loss = 0.0
metrics_score = [0.0, 0.0]
model.train()
for batch_data in tqdm(train_dataloader, desc=f'Training Epoch {epoch}/{num_epochs}', unit='epoch'):
inputs = batch_data["image"].float().to(device)
# print(inputs.shape)
labels = batch_data["seg"].float().to(device)
# print(labels.shape)
# print(torch.max(inputs))
# print(torch.min(inputs))
# print(torch.max(labels))
# print(torch.min(labels))
optimizer.zero_grad()
outputs = model(inputs)
# print(outputs.shape)
out_softmax = nn.Softmax(dim=1)(outputs)
# print(out_softmax.shape)
# print(torch.max(outputs))
# print(torch.min(outputs))
# print(outputs.shape)
# print(labels.shape)
train_loss = loss_function(out_softmax, labels)
# train_loss.requires_grad = True
train_loss.backward()
# print(outputs)
# print('------------------')
# print(labels)
score = calculate_metrics(out_softmax, labels)
metrics_score = list(map(add, metrics_score, score))
optimizer.step()
# print(train_loss.item())
# print(type(epoch_train_loss))
epoch_train_loss += train_loss.item()
epoch_train_loss = epoch_train_loss/len(train_dataloader)
epoch_train_jaccard = metrics_score[0]/len(train_dataloader)
epoch_train_acc = metrics_score[1]/len(train_dataloader)
wandb.log({
"train_loss":epoch_train_loss,
"train_jaccard":epoch_train_jaccard,
"train_acc":epoch_train_acc,
}, step=epoch)
overall_train_loss_per_epoch.append(train_loss.item())
overall_train_jaccard_per_epoch.append(epoch_train_jaccard)
overall_train_acc_per_epoch.append(epoch_train_acc)
model.eval()
metrics_score = [0.0, 0.0]
with torch.no_grad():
for batch_data in tqdm(val_dataloader, desc=f'Test Epoch {epoch}/{num_epochs}', unit='epoch'):
inputs = batch_data["image"].float().to(device)
labels = batch_data["seg"].float().to(device)
outputs = model(inputs)
# print(outputs.shape)
out_softmax = nn.Softmax(dim=1)(outputs)
# print(out_softmax.shape)
val_loss = loss_function(outputs, labels)
score = calculate_metrics(outputs, labels)
metrics_score = list(map(add, metrics_score, score))
optimizer.step()
# print(test_loss.item())
epoch_val_loss += val_loss.item()
epoch_val_loss = epoch_train_loss/len(val_dataloader)
epoch_val_jaccard = metrics_score[0]/len(val_dataloader)
epoch_val_acc = metrics_score[1]/len(val_dataloader)
torch.save({'epoch': epoch,
'model_state_dict': model.state_dict(),
'optimizer_state_dict': optimizer.state_dict()
},checkpoint_path)
wandb.log({
"val_loss":epoch_val_loss,
"val_jaccard":epoch_val_jaccard,
"val_acc":epoch_val_acc,
}, step=epoch)
# print(epoch_test_loss)
overall_val_loss_per_epoch.append(val_loss.item())
overall_val_jaccard_per_epoch.append(epoch_val_jaccard)
overall_val_acc_per_epoch.append(epoch_val_acc)
print(f'Epoch [{epoch + 1}/{num_epochs}], '
f'Train Loss: {train_loss.item():.4f}, '
f'Train Jaccard: {epoch_train_jaccard.item():.4f}, '
f'Train Accuracy: {epoch_train_acc.item():.4f}, '
f'Val Loss: {val_loss.item():.4f}, '
f'Val Jaccard: {epoch_val_jaccard.item():.4f}, '
f'Val Accuracy: {epoch_val_acc.item():.4f}, ')
if __name__=='__main__':
wandb.login()
wandb.init(project="3D-Liver-Segmentation",
config={"learning_rate":1e-5,
"epochs":20,
"num_classes":3,
"dropout":0.5,
"batch_size":1})
train_dataloader, val_dataloader, image_test, segmentation_test = get_dataloaders(wandb.config["batch_size"], wandb.config["num_classes"])
model, num_epochs,optimizer, loss, overall_train_loss_per_epoch, overall_train_jaccard_per_epoch, overall_train_acc_per_epoch, overall_test_loss_per_epoch, overall_test_jaccard_per_epoch, overall_test_acc_per_epoch = training_model(train_dataloader,val_dataloader,wandb.config["dropout"],wandb.config["epochs"], wandb.config["learning_rate"])
wandb.watch(model,loss,log="all", log_freq=10)
columns = ['image', 'segmentation']
df = pd.DataFrame(columns=columns)
for index in tqdm(range(len(image_test))):
df.loc[index, 'image'] = image_test[index]
df.loc[index, 'segmentation'] = segmentation_test[index]
df.to_csv('test_dataset.csv')