-
Notifications
You must be signed in to change notification settings - Fork 3
/
train.py
145 lines (117 loc) · 4.82 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
import torch
import torch.nn as nn
import numpy as np
import os
from models.unet_models import UNetResNet50_9, UNetResNet50_3
from torch.utils.tensorboard import SummaryWriter
from utils.dataloader import ThreeSeasonDataset
from torch.utils.data import DataLoader
from tqdm import tqdm
import yaml
import argparse
parser = argparse.ArgumentParser(description='Train script for the UNetResNet50_X model.')
parser.add_argument('--config', '-c', type=str, default='configs/config.yaml', help='Path to the config file.')
args = parser.parse_args()
# Load config file
with open(args.config) as f:
config = yaml.load(f, Loader=yaml.FullLoader)
log_dir = f'logs'
writer = SummaryWriter(log_dir)
# Hyperparameters
# Define Hyperparameters
num_epochs = config['training']['num_epochs']
batch_size = config['training']['batch_size']
learning_rate = config['training']['learning_rate']
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# Define model
model_name = config['model']['name']
save_name = config['model']['save_name']
model = globals()[model_name]()
model.to(device)
# Define loss function
criterion = nn.BCEWithLogitsLoss()
# Define optimizer
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
# Load data
dataset = globals()[config['data']['dataset']](root_dir=config['data']['train_path'])
# split the dataset into train and test and validation
train_size = int(0.8 * len(dataset))
test_size = len(dataset) - train_size
train_dataset, test_dataset = torch.utils.data.random_split(dataset, [train_size, test_size])
# split the train dataset into train and validation
train_size = int(0.8 * len(train_dataset))
val_size = len(train_dataset) - train_size
train_dataset, val_dataset = torch.utils.data.random_split(train_dataset, [train_size, val_size])
# Create dataloader
train_dataloader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
val_dataloader = DataLoader(val_dataset, batch_size=batch_size, shuffle=True)
test_dataloader = DataLoader(test_dataset, batch_size=batch_size, shuffle=True)
# Train model
total_step = len(train_dataloader)
for epoch in tqdm(range(num_epochs)):
model.train()
epoch_loss = 0
for i_batch, sample_batched in enumerate(train_dataloader):
# Get batch
images = sample_batched['image'].to(device)
filled = sample_batched['filled'].to(device)
border = sample_batched['border'].to(device)
# Forward pass
filled_pred, border_pred = model(images)
filled_loss = criterion(filled_pred, filled)
border_loss = criterion(border_pred, border)
loss = filled_loss + border_loss
# Backward pass
optimizer.zero_grad()
loss.backward()
optimizer.step()
# Print loss
if (i_batch + 1) % 10 == 0:
tqdm.write('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}'.format(epoch + 1, num_epochs, i_batch + 1, total_step, loss.item()))
# Log loss
writer.add_scalar('Loss/train', loss.item(), epoch * total_step + i_batch)
# Evaluate model
model.eval()
with torch.no_grad():
total_val_loss = 0
for i_batch, sample_batched in enumerate(val_dataloader):
# Get batch
images = sample_batched['image'].to(device)
filled = sample_batched['filled'].to(device)
border = sample_batched['border'].to(device)
# Forward pass
filled_pred, border_pred = model(images)
filled_loss = criterion(filled_pred, filled)
border_loss = criterion(border_pred, border)
loss = filled_loss + border_loss
total_val_loss += loss.item()
print('Epoch [{}/{}], Val Loss: {:.4f}'.format(epoch + 1, num_epochs, total_val_loss / len(val_dataloader)))
# Log loss
writer.add_scalar('Loss/val', total_val_loss / len(val_dataloader), epoch)
# Save the best model
if epoch == 0:
best_loss = total_val_loss
torch.save(model.state_dict(), f'logs/{save_name}.pt')
else:
if total_val_loss < best_loss:
best_loss = total_val_loss
torch.save(model.state_dict(), f'logs/{save_name}.pt')
# Test model
model.eval()
with torch.no_grad():
total_test_loss = 0
for i_batch, sample_batched in enumerate(test_dataloader):
# Get batch
images = sample_batched['image'].to(device)
filled = sample_batched['filled'].to(device)
border = sample_batched['border'].to(device)
# Forward pass
filled_pred, border_pred = model(images)
filled_loss = criterion(filled_pred, filled)
border_loss = criterion(border_pred, border)
loss = filled_loss + border_loss
total_test_loss += loss.item()
# Log loss
writer.add_scalar('Loss/test', total_test_loss / len(test_dataloader), epoch)
# Close Tensorboard writer
writer.close()