-
Notifications
You must be signed in to change notification settings - Fork 0
/
main_train_segmentation.py
173 lines (148 loc) · 6.74 KB
/
main_train_segmentation.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
"""
Main training script
"""
import pandas as pd
import os
import torch.nn as nn
import torch.backends.cudnn as cudnn
import torch.utils.data
from utils.utils import set_seed
from utils.segmentation_training_utils import validate, train
from datasets.skinSegmentationDatasetOptical import SkinSegmentationDataset, get_transforms_segmentation
from models.select_model import select_model
from datetime import datetime
import json
import segmentation_models_pytorch as smp
def main(config=None, s=1234):
with open('config_segmentation.json') as config_file:
paths = json.load(config_file)
ratio = paths['csvPath'].split('_')[2]
best_prec1 = 0
save_dir = paths['savePath']
cpu = False
resume = paths['resumePath']
start_epoch = 0
if cpu:
device = torch.device('cpu')
else:
device = torch.device('cuda')
cudnn.benchmark = True
set_seed(s)
# define loss function (criterion)
if config['loss'] == 'bce':
weights = torch.tensor(config['weights']).to(device)
criterion = nn.BCEWithLogitsLoss(pos_weight=weights)
else:
criterion = smp.losses.DiceLoss(mode=smp.losses.BINARY_MODE, from_logits=True)
criterion.to(device)
transforms_train, transforms_val = get_transforms_segmentation(config["img_size"], transform=config['transform'])
config['transforms'] = transforms_train.__str__()
# Check if the save_dir exists or not
if not os.path.exists(save_dir):
os.makedirs(save_dir)
df = pd.read_csv(paths['csvPath'])
folds = [1]
backbone = config['backbone']
for fold in folds:
# df_train = df
df_train = df[df['fold'] != fold]
df_valid = df[df['fold'] == fold]
print("train samples:", len(df_train), "validation sample:", len(df_valid))
# prepare model
model = select_model(config['net_name'], width=1, backbone=backbone)
print(model)
model.to(device)
# prepare optimizer
if config['optim'] == 'adam':
optimizer = torch.optim.Adam(model.parameters(), config['lr'],
weight_decay=config["weight_decay"])
else:
optimizer = torch.optim.SGD(model.parameters(), config["lr"],
momentum=config["momentum"],
weight_decay=config["weight_decay"])
scheduler_cosine = torch.optim.lr_scheduler.StepLR(optimizer, 1, gamma=1)
# optionally resume from a checkpoint
if resume:
if os.path.isfile(resume):
print("=> loading checkpoint '{}'".format(resume))
checkpoint = torch.load(resume)
start_epoch = checkpoint['epoch']
best_prec1 = checkpoint['best_prec1']
model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
print("=> loaded checkpoint (epoch {})"
.format(checkpoint['epoch']))
print(checkpoint.keys())
else:
print("=> no checkpoint found at '{}'".format(resume))
train_dataset = SkinSegmentationDataset(df_train,
img_path=paths['imagesPath'],
mask_path=paths['masksPath'],
transforms=transforms_train,
)
test_dataset = SkinSegmentationDataset(df_valid,
img_path=paths['imagesPath'],
mask_path=paths['masksPath'],
transforms=transforms_val,
)
batch_size = config["batch_size"]
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=config['num_workers'],
pin_memory=config['pin_memory'],
drop_last=True
)
val_loader = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=config['num_workers'],
pin_memory=config['pin_memory']
)
# visualisation of sample data
dt_string = datetime.now().strftime("%d-%m-%Y_%H-%M-%S")
for epoch in range(start_epoch, config["epochs"]):
train(train_loader, model, criterion, scheduler_cosine, optimizer, epoch, cpu)
# evaluate on validation set
prec1, _ = validate(val_loader, model, criterion, cpu)
# remember best prec@1 and save checkpoint
best_prec1 = max(prec1, best_prec1)
model_save_path = os.path.join(save_dir,
f'student_new_'
f'{dt_string}_{config["net_name"]}_{backbone}'
f'_optim={optimizer.__class__.__name__}'
f'_loss={config["loss"]}'
f'_e={epoch}'
f'_lr={config["lr"]}'
f'_bs={config["batch_size"]}'
f'_color={config["color_space"]}'
f'_w={config["weights"]}'
f'_{config["momentum"]}'
f'f={fold}_optical_s_r={ratio}_{s}.tar')
torch.save({
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
'best_prec1': best_prec1,
}, model_save_path)
if __name__ == '__main__':
cfg = {
"batch_size": 10,
"width": 1,
'img_size': 256,
'optim': 'sgd',
'lr': 0.006,
'momentum': 0.81,
'weight_decay': 0.02,
'epochs': 65,
'net_name': 'unet',
'num_workers': 0,
'pin_memory': True,
"weights": [1.25],
'loss': 'dice',
"backbone": "resnet18",
'transform': "optical"
}
seeds = [2137]
for seed in seeds:
main(cfg, seed)