FFC3DLungDIR/train.py at main · mohammadhashemii/FFC3DLungDIR · GitHub

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
import argparse
import torch
from torch.utils.data import DataLoader
from dataset import CREATISDataset, DIRLABDataset
from utils.yaml_reader import load_config
from models.model import FFCResNetGenerator
from models.SpatialTransformerNetwork import SpatialTransformation
from losses import compute_loss
from tqdm import tqdm
from utils.view3d_image import *
import re

parser = argparse.ArgumentParser()
parser.add_argument('--training_config_path', type=str, default='./configs/training_settings.yaml', help='path to the training config')
parser.add_argument('--model_config_path', type=str, default='./configs/FFCResnetGenerator_settings.yaml', help='path to the model config')
parser.add_argument('--exp', type=str, required=True, help='experiment number')
parser.add_argument('--seed', type=int, default=23)
parser.add_argument('--starting_epoch', type=int)


args = parser.parse_args()

torch.manual_seed(args.seed)

# Loading configs
train_configs = load_config(args.training_config_path)
model_configs = load_config(args.model_config_path)

print(train_configs)
print(model_configs)

# Creating datasets
train_dataset = CREATISDataset(root=train_configs['train_data_path'],
                            case_list=train_configs['train_cases'])

val_dataset = CREATISDataset(root=train_configs['train_data_path'],
                            case_list=train_configs['val_cases'])


print('TRAIN DATA: {} pairs of fixed/moving images.'.format(len(train_dataset)))
print('VALIDATION DATA: {} pairs of fixed/moving images.'.format(len(val_dataset)))

train_loader = DataLoader(train_dataset, **train_configs['data_loader'])
val_loader = DataLoader(val_dataset, **train_configs['data_loader'])

# Model defining
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
print(f"Using device: {device}")
FFCGenerator = FFCResNetGenerator(input_nc=model_configs['input_nc'],
                                  output_nc=model_configs['output_nc'],
                                  n_downsampling=model_configs['n_downsampling'],
                                  n_blocks=model_configs['n_blocks'],
                                  init_conv_kwargs=model_configs['init_conv_kwargs'],
                                  downsample_conv_kwargs=model_configs['downsample_conv_kwargs'],
                                  resnet_conv_kwargs=model_configs['resnet_conv_kwargs']).to(device)
stn = SpatialTransformation(use_gpu=False)


# optimization settings
optimizer = torch.optim.Adam(FFCGenerator.parameters(),
                            lr=train_configs['lr'])


# Load the weights
if args.starting_epoch is not None:
    generator_weights = "saved_ours/exp" + args.exp + "_epoch" + str(args.starting_epoch) + "_gen.pth"
    stn_weights = "saved_ours/exp" + args.exp + "_epoch" + str(args.starting_epoch) + "_stn.pth"

    FFCGenerator.load_state_dict(torch.load(generator_weights))
    stn.load_state_dict(torch.load(stn_weights))

if args.starting_epoch is None:
    starting_epoch = 0

# Loop over epochs for training
for epoch in range(starting_epoch, starting_epoch + train_configs['n_epochs'] + 1):

    batch_counter = 0
    # epoch losses
    epoch_train_loss = 0.0
    epoch_train_ncc = 0.0
    epoch_train_sm = 0.0

    for paired_patches in train_loader:
        batch_counter += 1
        # paired_patches shape: [batch, 2, d, h, w]
        pair = paired_patches.to(device)
        DVF = FFCGenerator(pair).cpu()

        fi = torch.unsqueeze(pair[:, 0, :], 1).cpu() # fixed patch
        mi = torch.unsqueeze(pair[:, 1, :], 1).cpu() # moving patch
        registered_images = stn(mi.permute(0, 1, 4, 3, 2), # (batch, c, w, h, d)
                                DVF.permute(0, 1, 4, 3, 2))

        registered_images = registered_images.permute(0, 1, 4, 3, 2)
        # compute loss
        loss, ncc, sm = compute_loss(y_pred=registered_images.to(device),
                                y_true=fi.to(device),
                                dvf=DVF.to(device),
                                window_size=train_configs['window_size'],
                                lamda=train_configs['lamda'],
                                mu1=train_configs['mu1'],
                                mu2=train_configs['mu2'])

        epoch_train_loss += loss.item()
        epoch_train_ncc += ncc.item()
        epoch_train_sm += sm.item()

        optimizer.zero_grad()

        loss.backward()
        optimizer.step()

    if epoch % train_configs['save_every'] == 0:
        print("Epoch: {}, train_loss={}, train_ncc_loss={}, train_smoothing_loss= {}".format(epoch, epoch_train_loss / len(train_loader),
                                                                            epoch_train_ncc / len(train_loader),
                                                                            epoch_train_sm / len(train_loader)))


        # validation
        with torch.no_grad():
            # epoch losses
            epoch_val_loss = 0.0
            epoch_val_ncc = 0.0
            epoch_val_sm = 0.0
            for paired_patches in val_loader:

                # paired_patches shape: [batch, 2, d, h, w]
                pair = paired_patches.to(device)
                DVF = FFCGenerator(pair).cpu()

                fi = torch.unsqueeze(pair[:, 0, :], 1).cpu() # fixed patch
                mi = torch.unsqueeze(pair[:, 1, :], 1).cpu() # moving patch
                registered_images = stn(mi.permute(0, 1, 4, 3, 2), # (batch, c, w, h, d)
                                        DVF.permute(0, 1, 4, 3, 2))

                registered_images = registered_images.permute(0, 1, 4, 3, 2)

                # compute loss
                loss, ncc, sm = compute_loss(y_pred=registered_images.to(device),
                                        y_true=fi.to(device),
                                        dvf=DVF.to(device),
                                        window_size=train_configs['window_size'],
                                        lamda=train_configs['lamda'],
                                        mu1=train_configs['mu1'],
                                        mu2=train_configs['mu2'])

                epoch_val_loss += loss.item()
                epoch_val_ncc += ncc.item()
                epoch_val_sm += sm.item()

            print("val_loss={}, val_ncc_loss={}, val_smoothing_loss={}".format(epoch_val_loss / len(val_loader),
                                                                            epoch_val_ncc / len(val_loader),
                                                                            epoch_val_sm / len(val_loader)))

        torch.save(FFCGenerator.state_dict(), train_configs['save_dir'] + "exp" + args.exp + "_epoch" + str(epoch) + "_gen.pth")
        torch.save(stn.state_dict(), train_configs['save_dir'] + "exp" + args.exp + "_epoch" + str(epoch) + "_stn.pth")

    # if batch_counter % train_configs['save_every'] == 0:
    #     torch.save(FFCGenerator.state_dict(), train_configs['save_dir'] + "exp" + str(EXPERIMENT) + "_patch64_generator.pth")
    #     torch.save(stn.state_dict(), train_configs['save_dir'] + "exp" + str(EXPERIMENT) + "_patch64_stn.pth")

    #     for c in range(10):
    #         print("--------------")
    #         test_per_case(experiment=EXPERIMENT, case_num=c+1, phases_list=[0, 5], epoch_idx=epoch)