-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
143 lines (114 loc) · 5.35 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
import torch
import torch.nn as nn
import torch.optim as optim
import numpy as np
import pprint
from tqdm import tqdm
from utility.log import log_terminal
from utility.train import set_parameters, rmse, geom_element, angle_element
from utility.visualization import visualize
def train_function(args, DEVICE, model, loss_fn_pixel, optimizer, train_loader):
total_loss = 0
model.train()
for image, label, _, _, label_list in tqdm(train_loader):
image = image.to(device=DEVICE)
label = label.float().to(device=DEVICE)
prediction = model(image)
loss = loss_fn_pixel(prediction, label) * args.pixel_loss_weight
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_loss += loss.item()
return total_loss
def validate_function(args, DEVICE, model, epoch, val_loader):
print("=====Starting Validation=====")
model.eval()
dice_score, rmse_total = 0, 0
extracted_pixels_list = []
rmse_list = [[0]*len(val_loader) for _ in range(args.output_channel)]
angle_list = [[0]*len(val_loader) for _ in range(len(args.label_for_angle))]
with torch.no_grad():
for idx, (image, label, image_path, image_name, label_list) in enumerate(tqdm(val_loader)):
image = image.to(DEVICE)
label = label.to(DEVICE)
image_path = image_path[0]
image_name = image_name[0].split('.')[0]
prediction = model(image)
# validate angle difference
if args.label_for_angle != []:
predict_angle, label_angle = angle_element(args, prediction, label_list, DEVICE)
for i in range(len(args.label_for_angle)):
angle_list[i][idx] = abs(label_angle[i] - predict_angle[i])
# validate mean geom difference
predict_spatial_mean, label_spatial_mean = geom_element(torch.sigmoid(prediction), label)
## get rmse difference
rmse_list, index_list = rmse(args, prediction, label_list, idx, rmse_list)
extracted_pixels_list.append(index_list)
## make predictions to be 0. or 1.
prediction_binary = (prediction > 0.5).float()
dice_score += (2 * (prediction_binary * label).sum()) / ((prediction_binary + label).sum() + 1e-8)
## visualize
if epoch % args.dilation_epoch == 0 or epoch % args.dilation_epoch == (args.dilation_epoch-1):
if not args.no_visualization:
visualize(
args, idx, image_path, image_name, label_list, epoch, extracted_pixels_list, prediction, prediction_binary,
predict_spatial_mean, label_spatial_mean, None, 'train'
)
dice = dice_score/len(val_loader)
# Removing RMSE for annotation that does not exist in the label
rmse_mean_by_label = []
for i in range(len(rmse_list)):
tmp_sum, count = 0, 0
for j in range(len(rmse_list[i])):
if rmse_list[i][j] != -1:
tmp_sum += rmse_list[i][j]
count += 1
rmse_mean_by_label.append(tmp_sum/count)
total_rmse_mean = sum(rmse_mean_by_label)/len(rmse_mean_by_label)
print(f"Dice score: {dice}")
print(f"Average Pixel to Pixel Distance: {total_rmse_mean}")
if args.label_for_angle != []:
# add up angle values
angle_value = []
for i in range(len(args.label_for_angle)):
angle_value.append(sum(angle_list[i]))
angle_value.append(sum(list(map(sum, angle_list))))
return dice, total_rmse_mean, rmse_list, rmse_mean_by_label, angle_value
else:
return dice, total_rmse_mean, rmse_list, rmse_mean_by_label, 0
def train(args, model, DEVICE):
best_loss, best_rmse_mean, best_angle_diff = np.inf, np.inf, np.inf
optimizer = optim.Adam(model.parameters(), lr=args.lr)
for epoch in range(args.epochs):
print(f"\nRunning Epoch # {epoch}")
if epoch % args.dilation_epoch == 0:
args, loss_fn, train_loader, val_loader = set_parameters(
args, model, epoch, DEVICE
)
loss = train_function(
args, DEVICE, model, loss_fn, optimizer, train_loader
)
dice, rmse_mean, rmse_list, rmse_mean_by_label, angle_value = validate_function(
args, DEVICE, model, epoch, val_loader
)
print("Average Train Loss: ", loss/len(train_loader))
if best_loss > loss:
print("=====New best model=====")
best_loss = loss
if best_rmse_mean > rmse_mean:
checkpoint = {
"state_dict": model.state_dict(),
"optimizer": optimizer.state_dict(),
}
torch.save(checkpoint, f'./results/{args.wandb_name}/best.pth')
best_rmse_mean = rmse_mean
best_rmse_list = rmse_list
if args.label_for_angle != []:
if best_angle_diff > angle_value[len(args.label_for_angle)]:
checkpoint = {
"state_dict": model.state_dict(),
"optimizer": optimizer.state_dict(),
}
torch.save(checkpoint, f'./results/{args.wandb_name}/best_angle.pth')
best_angle_diff = angle_value[len(args.label_for_angle)]
log_terminal(args, 'best_rmse', best_rmse_list)