GroupMorph/train.py at main · TVayne/GroupMorph · 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
import os, glob, sys, datetime
from argparse import ArgumentParser
import torch.utils.data as Data
from torch.optim import lr_scheduler
from Model.Loss import *
from Model.Net import GruopMorph
from Model.Function import Dataset_OASIS, SpatialTransformer

parser = ArgumentParser()
parser.add_argument("--lr", type=float,
                    dest="lr", default=1e-4, help="learning rate")
parser.add_argument("--iteration", type=int,
                    dest="iteration", default=160001,
                    help="number of total iterations")
parser.add_argument("--local_ori", type=float,
                    dest="local_ori", default=0,
                    help="Local Orientation Consistency loss: suggested range 1 to 1000")
parser.add_argument("--smooth", type=float,
                    dest="smooth", default=1,
                    help="Gradient smooth loss: suggested range 0.1 to 10")
parser.add_argument("--dice", type=float,
                    dest="dice", default=1,
                    help="Dice loss: suggested range 0.1 to 10")
parser.add_argument("--checkpoint", type=int,
                    dest="checkpoint", default=4000,
                    help="frequency of saving models")
parser.add_argument("--bs_ch", type=int,
                    dest="bs_ch", default=8,
                    help="number of basic channels")
parser.add_argument("--modelname", type=str,
                    dest="model_name",
                    default='reg',
                    help="Name for saving")
parser.add_argument("--gpu", type=str,
                    dest="gpu",
                    default='0',
                    help="gpus")
parser.add_argument("--classes", type=int,
                    dest="classes",
                    default='36',
                    help="number classes")
opt = parser.parse_args()

lr = opt.lr
bs_ch = opt.bs_ch
local_ori = opt.local_ori
n_checkpoint = opt.checkpoint
smooth = opt.smooth
dice = opt.dice
model_name = opt.model_name
iteration = opt.iteration
classes = opt.classes
os.environ["CUDA_VISIBLE_DEVICES"] = opt.gpu


imgshape = (160, 192, 192)
groups = (4, 2, 2)  # (4,4,4), (4,4,2), (4,2,2) or (2,2,2)
def train():

    model = GruopMorph(1, 8, imgshape, groups).cuda()


    loss_similarity = ncc_loss
    transfor = SpatialTransformer().cuda()
    optimizer = torch.optim.Adam(model.parameters(), lr=lr)
    scheduler = lr_scheduler.MultiStepLR(optimizer, milestones=[88000], gamma=0.1)

    model_dir = 'Model_weight/T-test1'
    if not os.path.isdir(model_dir):
        os.mkdir(model_dir)

    loss_all = np.zeros((5, iteration + 1))

    step = 0

    load_model = False
    if load_model is True:
        model_path = ''
        step = 88000
        model.load_state_dict(torch.load(model_path))
        loss_load = np.load("")
        loss_all[:, :step] = loss_load[:, :step]

    names = sorted(glob.glob('../neurite-oasis.v1.0/OASIS_OAS1_*_MR1'))[0:255]  # 255
    training_generator = Data.DataLoader(Dataset_OASIS(names, norm=False), batch_size=1,
                                         shuffle=True, num_workers=2)
    fixed_names = sorted(glob.glob('../neurite-oasis.v1.0/OASIS_OAS1_*_MR1'))[277:282]  # 255
    fixed_generator = Data.DataLoader(Dataset_OASIS(fixed_names, norm=False), batch_size=1,
                                         shuffle=True, num_workers=2)

    while step <= iteration:
        for batch_idx, data in enumerate(training_generator):
            X, X_label = data['image'].cuda(), data['image_label'].cuda()
            for batch_idx_fixed, data_fixed in enumerate(fixed_generator):
                Y, Y_label = data_fixed['image'].cuda(), data_fixed['image_label'].cuda()
                X = X.float()
                Y = Y.float()

                flows, warps, smo = model(X, Y)

                # dice loss
                Y_label_onehot = mask_to_one_hot(Y_label, n_classes=classes)
                X_label_onehot = mask_to_one_hot(X_label, n_classes=classes)
                warps_label_onehot = transfor(X_label_onehot, flows)
                diceloss = compute_per_channel_dice(warps_label_onehot, Y_label_onehot, classes=classes)
                # dice loss

                sim = loss_similarity(warps, Y)
                smo_loss = smo
                loss = sim + dice * diceloss + smooth * smo_loss
                optimizer.zero_grad()
                loss.backward()

                optimizer.step()
                scheduler.step()

                l = optimizer.param_groups[0]['lr']
                loss_all[:, step] = np.array(
                    [loss.item(), sim.item(), diceloss.item(), smo_loss.item(), sim.item()])
                sys.stdout.write(
                    "\r" + 'step "{0}" -> training loss "{1:.4f}" - sim_loss "{2:4f}" - dice_loss "{3:4f}" - smo_loss "{4:4f}" - lr:"{5:.6f}"'.format(
                        step, loss.item(), sim.item(), diceloss.item(), smo_loss.item(), l))
                sys.stdout.flush()
                # log_dir = "log/loss.txt"
                # with open(log_dir, "a") as log:
                #     log.write(
                #     "\n" + 'step "{0}" -> training loss "{1:.4f}" - sim_loss "{2:4f}" - dice_loss "{3:4f}" - smo_loss "{4:4f}" - lr:"{5:.6f}"'.format(
                #         step, loss.item(), sim.item(), diceloss.item(), smo_loss.item(), l))

                step += 1
                if (step % n_checkpoint == 0):
                    # save model
                    modelname = model_dir + '/' + model_name + str(
                        step) + '.pth'
                    torch.save(model.state_dict(), modelname)
                    np.save(model_dir + '/loss' + model_name + str(step) + '.npy', loss_all)
                    # save model
                    valname = sorted(glob.glob('../neurite-oasis.v1.0/OASIS_OAS1_*_MR1'))[255:277]
                    valid_generator = Data.DataLoader(Dataset_OASIS(valname, norm=False),
                                                      batch_size=1,
                                                      shuffle=False, num_workers=2)

                    dice_total = []
                    print("\nValiding...")
                    for batch_idx, data in enumerate(valid_generator):
                        X, X_label = data['image'].cuda(), data['image_label'].cuda()
                        for batch_idx_fixed, data_fixed in enumerate(fixed_generator):
                            Y, Y_label = data_fixed['image'].cuda(), data_fixed['image_label'].cuda()
                            with torch.no_grad():
                                X = X.float()
                                Y = Y.float()
                                flows, _, _ = model(X, Y)
                                X_Y_label = transfor(X_label, flows, mode='nearest')[0, 0, :, :, :]
                                dice_score_reg = dicegup(X_Y_label, Y_label[0, 0, :, :, :])
                                dice_score_reg = dice_score_reg.cpu().numpy()
                                dice_total.append(dice_score_reg)
                    dice_total = np.array(dice_total)
                    print("Dice mean: ", dice_total.mean())
                    log_dir = "log/val.txt"
                    with open(log_dir, "a") as log:
                        log.write("step:" + str(step) + "Dice mean:" + str(dice_total.mean()) + "\n")
                if step > iteration:
                    break


if __name__ == '__main__':
    start = datetime.datetime.now()
    train()
    end = datetime.datetime.now()
    print("Time used:", end - start)