-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
176 lines (143 loc) · 5.59 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
from typing import List
import numpy as np
import matplotlib.pyplot as plt
from tqdm import tqdm
from omegaconf import OmegaConf
from segmentation_models_pytorch.losses import DiceLoss # for image segmentation task
import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torchsummary import summary
from segment.helper import show_image, plot_loss
from segment.nn_models import UNet
from segment.datasets import get_data_set
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# ----- Parameters
cfg = OmegaConf.load("conf/config.yaml")
# Data
data_set : str = cfg.trainer.data_set
# Trainer
epochs: int = cfg.trainer.epochs
batch_size: int = cfg.trainer.batch_size
test_size: float = cfg.trainer.test_size
# optimizer parameters
amp: bool = cfg.optim.amp
gradient_clipping: float = cfg.optim.gradient_clipping
optim_type: str = cfg.optim.optim_type
learning_rate: float = cfg.optim.learning_rate
weight_decay: float = cfg.optim.weight_decay
momentum: float = cfg.optim.momentum
# Unet parameters
bilinear: bool = cfg.model.bilinear
unet_base_exp: int = cfg.model.unet_base_exp
unet_dims: List = []
for i in range(5):
unet_dims.append(2**(unet_base_exp + i))
# ----- Loading data into dataloaders
X_train, X_test = get_data_set(data_set=data_set, test_size=test_size)
print(f"Size of Train Dataset : {len(X_train)}")
print(f"Size of Valid Dataset : {len(X_test)}")
train_loader = DataLoader(dataset=X_train,
batch_size=batch_size,
shuffle=True,
num_workers=2,
pin_memory=True)
test_loader = DataLoader(dataset=X_test,
batch_size=batch_size,
shuffle=False,
num_workers=2)
print(f"num of batches in train: {len(train_loader)}, and test: {len(test_loader)}")
# ----- instantiate model
model = UNet(n_channels=3,
n_classes=1,
bilinear=bilinear,
ddims=unet_dims,
UQ=True,
)
model.to(device)
summary(model, X_train[0][0].shape)
# ----- Loss criterion
def compose_loss_fn(n_classes: int = 1,
alpha: float = 1.0,
beta: float = 1.0,):
"""The overall loss function is a linear combination of BCE/CE and Dice"""
criterion = nn.CrossEntropyLoss() if n_classes > 1 else nn.BCEWithLogitsLoss()
dice_loss = DiceLoss(mode='binary')
total = 0.5 * (alpha + beta)
alpha /= total
beta /= total
def loss_fn(masks_pred, masks):
loss = alpha * criterion(masks_pred, masks.float())
loss += beta * dice_loss(masks_pred, masks.float())
return loss
return loss_fn
loss_fn = compose_loss_fn(n_classes=model.n_classes,
alpha=cfg.model.loss.bce_frac,
beta=cfg.model.loss.dice_frac)
# ----- Optimizer + Scheduler + AMP
if optim_type=="adam":
optimizer = torch.optim.Adam(model.parameters(),
lr=learning_rate,
weight_decay=weight_decay,)
elif optim_type=="rmsprop":
optimizer = torch.optim.RMSprop(model.parameters(),
lr=learning_rate,
weight_decay=weight_decay,
momentum=momentum,
foreach=True,)
elif optim_type=="sgd":
optimizer = torch.optim.SGD(model.parameters(),
lr=learning_rate,
momentum=momentum)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, 'min', patience=5)
grad_scaler = torch.cuda.amp.GradScaler(enabled=amp)
# ----- Training Loop
train_losses = []
test_losses = []
best_test_loss= np.Inf
for i in range(epochs):
# ---- train
train_epoch_loss = 0.0
for images, masks in tqdm(train_loader):
images = images.to(device)
masks = masks.to(device)
with torch.autocast(device.type if device.type != 'mps' else 'cpu', enabled=amp):
masks_pred = model(images)
loss = loss_fn(masks_pred, masks)
train_epoch_loss += loss / len(train_loader)
optimizer.zero_grad(set_to_none=True)
grad_scaler.scale(loss).backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), gradient_clipping)
grad_scaler.step(optimizer)
grad_scaler.update()
# ---- test
with torch.no_grad():
test_epoch_loss = 0.0
for images, masks in tqdm(test_loader):
images = images.to(device)
masks = masks.to(device)
masks_pred = model(images)
t_loss = loss_fn(masks_pred, masks)
test_epoch_loss += t_loss / len(test_loader)
scheduler.step(test_epoch_loss)
if test_epoch_loss < best_test_loss:
torch.save(model.state_dict(), 'best_model.pt')
print('MODEL SAVED')
best_test_loss = test_epoch_loss
# ---- logging
train_losses.append(train_epoch_loss.detach().cpu().numpy())
test_losses.append(test_epoch_loss.detach().cpu().numpy())
print(f'Epoch:{i}, train_loss: {train_epoch_loss}, test_loss: {test_epoch_loss}')
# ----- Inference
model.load_state_dict(torch.load('best_model.pt'))
idx = 5
image, mask = X_test[idx]
out = model(image.to(device).unsqueeze(0)) # (C,H,W) --> (1,C,H,W) to add batch dim
out = torch.sigmoid(out)
out = (out > 0.5) * 0.1
show_image(image, mask, out.detach().cpu().squeeze(0))
plt.savefig('test.png')
plt.close()
plot_loss(np.arange(epochs), train_losses, test_losses)
plt.savefig('losses.png')
plt.close()