Skip to content

Commit 410cb08

Browse files
junqiangchenjunqiangchen
authored
Add files via upload
1 parent 6104d40 commit 410cb08

22 files changed

+836
-48
lines changed

‎dataprocess/Augmain.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
from dataprocess.Augmentation.ImageAugmentation import DataAug3D
22

33
if __name__ == '__main__':
4-
aug = DataAug3D(rotation=5, width_shift=0.01, height_shift=0.01, depth_shift=0, zoom_range=0,
4+
aug = DataAug3D(rotation=10, width_shift=0.01, height_shift=0.01, depth_shift=0, zoom_range=0,
55
vertical_flip=True, horizontal_flip=True)
6-
aug.DataAugmentation('data/traindata.csv', 15, aug_path='D:\challenge\data\KiPA2022\\trainstage\\augtrain/')
6+
aug.DataAugmentation('data/traindata.csv', 10, aug_path='D:\challenge\data\KiPA2022\\trainstage\\augtrain/')

‎dataprocess/__pycache__/utils.cpython-36.pyc

6.71 KB
Binary file not shown.

‎dataprocess/data/trainaugdata.csv

Lines changed: 651 additions & 0 deletions
Large diffs are not rendered by default.

‎dataprocess/data/traindata.csv

Lines changed: 66 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,66 @@
1+
Image,Mask
2+
D:\challenge\data\KiPA2022\trainstage\train/Image/0.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/0.npy
3+
D:\challenge\data\KiPA2022\trainstage\train/Image/1.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/1.npy
4+
D:\challenge\data\KiPA2022\trainstage\train/Image/10.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/10.npy
5+
D:\challenge\data\KiPA2022\trainstage\train/Image/11.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/11.npy
6+
D:\challenge\data\KiPA2022\trainstage\train/Image/12.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/12.npy
7+
D:\challenge\data\KiPA2022\trainstage\train/Image/13.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/13.npy
8+
D:\challenge\data\KiPA2022\trainstage\train/Image/14.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/14.npy
9+
D:\challenge\data\KiPA2022\trainstage\train/Image/15.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/15.npy
10+
D:\challenge\data\KiPA2022\trainstage\train/Image/16.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/16.npy
11+
D:\challenge\data\KiPA2022\trainstage\train/Image/17.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/17.npy
12+
D:\challenge\data\KiPA2022\trainstage\train/Image/18.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/18.npy
13+
D:\challenge\data\KiPA2022\trainstage\train/Image/19.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/19.npy
14+
D:\challenge\data\KiPA2022\trainstage\train/Image/2.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/2.npy
15+
D:\challenge\data\KiPA2022\trainstage\train/Image/20.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/20.npy
16+
D:\challenge\data\KiPA2022\trainstage\train/Image/21.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/21.npy
17+
D:\challenge\data\KiPA2022\trainstage\train/Image/22.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/22.npy
18+
D:\challenge\data\KiPA2022\trainstage\train/Image/23.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/23.npy
19+
D:\challenge\data\KiPA2022\trainstage\train/Image/24.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/24.npy
20+
D:\challenge\data\KiPA2022\trainstage\train/Image/25.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/25.npy
21+
D:\challenge\data\KiPA2022\trainstage\train/Image/26.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/26.npy
22+
D:\challenge\data\KiPA2022\trainstage\train/Image/27.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/27.npy
23+
D:\challenge\data\KiPA2022\trainstage\train/Image/28.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/28.npy
24+
D:\challenge\data\KiPA2022\trainstage\train/Image/29.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/29.npy
25+
D:\challenge\data\KiPA2022\trainstage\train/Image/3.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/3.npy
26+
D:\challenge\data\KiPA2022\trainstage\train/Image/30.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/30.npy
27+
D:\challenge\data\KiPA2022\trainstage\train/Image/31.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/31.npy
28+
D:\challenge\data\KiPA2022\trainstage\train/Image/32.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/32.npy
29+
D:\challenge\data\KiPA2022\trainstage\train/Image/33.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/33.npy
30+
D:\challenge\data\KiPA2022\trainstage\train/Image/34.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/34.npy
31+
D:\challenge\data\KiPA2022\trainstage\train/Image/35.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/35.npy
32+
D:\challenge\data\KiPA2022\trainstage\train/Image/36.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/36.npy
33+
D:\challenge\data\KiPA2022\trainstage\train/Image/37.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/37.npy
34+
D:\challenge\data\KiPA2022\trainstage\train/Image/38.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/38.npy
35+
D:\challenge\data\KiPA2022\trainstage\train/Image/39.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/39.npy
36+
D:\challenge\data\KiPA2022\trainstage\train/Image/4.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/4.npy
37+
D:\challenge\data\KiPA2022\trainstage\train/Image/40.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/40.npy
38+
D:\challenge\data\KiPA2022\trainstage\train/Image/41.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/41.npy
39+
D:\challenge\data\KiPA2022\trainstage\train/Image/42.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/42.npy
40+
D:\challenge\data\KiPA2022\trainstage\train/Image/43.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/43.npy
41+
D:\challenge\data\KiPA2022\trainstage\train/Image/44.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/44.npy
42+
D:\challenge\data\KiPA2022\trainstage\train/Image/45.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/45.npy
43+
D:\challenge\data\KiPA2022\trainstage\train/Image/46.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/46.npy
44+
D:\challenge\data\KiPA2022\trainstage\train/Image/47.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/47.npy
45+
D:\challenge\data\KiPA2022\trainstage\train/Image/48.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/48.npy
46+
D:\challenge\data\KiPA2022\trainstage\train/Image/49.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/49.npy
47+
D:\challenge\data\KiPA2022\trainstage\train/Image/5.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/5.npy
48+
D:\challenge\data\KiPA2022\trainstage\train/Image/50.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/50.npy
49+
D:\challenge\data\KiPA2022\trainstage\train/Image/51.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/51.npy
50+
D:\challenge\data\KiPA2022\trainstage\train/Image/52.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/52.npy
51+
D:\challenge\data\KiPA2022\trainstage\train/Image/53.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/53.npy
52+
D:\challenge\data\KiPA2022\trainstage\train/Image/54.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/54.npy
53+
D:\challenge\data\KiPA2022\trainstage\train/Image/55.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/55.npy
54+
D:\challenge\data\KiPA2022\trainstage\train/Image/56.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/56.npy
55+
D:\challenge\data\KiPA2022\trainstage\train/Image/57.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/57.npy
56+
D:\challenge\data\KiPA2022\trainstage\train/Image/58.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/58.npy
57+
D:\challenge\data\KiPA2022\trainstage\train/Image/59.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/59.npy
58+
D:\challenge\data\KiPA2022\trainstage\train/Image/6.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/6.npy
59+
D:\challenge\data\KiPA2022\trainstage\train/Image/60.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/60.npy
60+
D:\challenge\data\KiPA2022\trainstage\train/Image/61.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/61.npy
61+
D:\challenge\data\KiPA2022\trainstage\train/Image/62.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/62.npy
62+
D:\challenge\data\KiPA2022\trainstage\train/Image/63.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/63.npy
63+
D:\challenge\data\KiPA2022\trainstage\train/Image/64.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/64.npy
64+
D:\challenge\data\KiPA2022\trainstage\train/Image/7.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/7.npy
65+
D:\challenge\data\KiPA2022\trainstage\train/Image/8.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/8.npy
66+
D:\challenge\data\KiPA2022\trainstage\train/Image/9.npy,D:\challenge\data\KiPA2022\trainstage\train/Mask/9.npy

‎dataprocess/data/validata.csv

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,6 @@
1+
Image,Mask
2+
D:\challenge\data\KiPA2022\trainstage\validation/Image/0.npy,D:\challenge\data\KiPA2022\trainstage\validation/Mask/0.npy
3+
D:\challenge\data\KiPA2022\trainstage\validation/Image/1.npy,D:\challenge\data\KiPA2022\trainstage\validation/Mask/1.npy
4+
D:\challenge\data\KiPA2022\trainstage\validation/Image/2.npy,D:\challenge\data\KiPA2022\trainstage\validation/Mask/2.npy
5+
D:\challenge\data\KiPA2022\trainstage\validation/Image/3.npy,D:\challenge\data\KiPA2022\trainstage\validation/Mask/3.npy
6+
D:\challenge\data\KiPA2022\trainstage\validation/Image/4.npy,D:\challenge\data\KiPA2022\trainstage\validation/Mask/4.npy

‎dataprocess/data/validata1.csv

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,2 @@
1+
Image,Mask
2+
D:\challenge\data\KiPA2022\trainstage\validation/Image/0.npy,D:\challenge\data\KiPA2022\trainstage\validation/Mask/0.npy

‎dataprocess/说明文档.txt

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,4 @@
1+
1、统计数据平均大小mean size,mean spacing: (array([153.29230769, 153.29230769, 197.49230769]), array([0.63416487, 0.63416487, 0.63416487]))
2+
2、图像缩放到固定大小（112x112x128）
3+
3、归一化采用（5，95）归一化范围
4+
4、损失采用focalloss

‎inference.py

Lines changed: 30 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,30 @@
1+
import torch
2+
import os
3+
from model import *
4+
from dataprocess.utils import file_name_path
5+
import SimpleITK as sitk
6+
7+
# Use CUDA
8+
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
9+
os.environ['CUDA_LAUNCH_BLOCKING'] = '1'
10+
use_cuda = torch.cuda.is_available()
11+
12+
13+
def inferencemutilunet3dtest():
14+
newSize = (112, 112, 128)
15+
Unet3d = MutilUNet3dModel(image_depth=128, image_height=112, image_width=112, image_channel=1, numclass=1,
16+
batch_size=1, loss_name='MutilFocalLoss', inference=True,
17+
model_path=r'log\MutilUNet3d\focalloss\BinaryVNet2dSegModel.pth')
18+
datapath = r"F:\MedicalData\(ok)2022KiPA\dataset\test\image"
19+
makspath = r"F:\MedicalData\(ok)2022KiPA\dataset\test\label"
20+
image_path_list = file_name_path(datapath, False, True)
21+
for i in range(len(image_path_list)):
22+
imagepathname = datapath + "/" + image_path_list[i]
23+
sitk_image = sitk.ReadImage(imagepathname)
24+
sitk_mask = Unet3d.inference(sitk_image, newSize)
25+
maskpathname = makspath + "/" + image_path_list[i]
26+
sitk.WriteImage(sitk_mask, maskpathname)
27+
28+
29+
if __name__ == '__main__':
30+
inferencemutilunet3dtest()

‎model/__pycache__/losses.cpython-36.pyc

-610 Bytes
Binary file not shown.

‎model/__pycache__/metric.cpython-36.pyc

651 Bytes
Binary file not shown.

‎model/__pycache__/modelResNet.cpython-36.pyc

245 Bytes
Binary file not shown.

‎model/__pycache__/modelUnet.cpython-36.pyc

222 Bytes
Binary file not shown.

‎model/__pycache__/modelVNet.cpython-36.pyc

102 Bytes
Binary file not shown.

‎model/losses.py

Lines changed: 2 additions & 18 deletions
Original file line numberDiff line numberDiff line change
@@ -202,7 +202,7 @@ def forward(self, y_pred_logits, y_true):
202202
intersection = torch.sum(y_true * y_pred, dim=(0, 2))
203203
denominator = torch.sum(y_true + y_pred, dim=(0, 2))
204204
gen_dice_coef = ((2. * intersection + smooth) / (denominator + smooth)).clamp_min(eps)
205-
loss = 1 - gen_dice_coef
205+
loss = - gen_dice_coef
206206
# Dice loss is undefined for non-empty classes
207207
# So we zero contribution of channel that does not have true pixels
208208
# NOTE: A better workaround would be to use loss term `mean(y_pred)`
@@ -212,27 +212,11 @@ def forward(self, y_pred_logits, y_true):
212212
return (loss * self.alpha).mean()
213213

214214

215-
class MutilCrossEntropyDiceLoss(nn.Module):
216-
"""
217-
mutil ce and dice
218-
"""
219-
220-
def __init__(self, alpha):
221-
super(MutilCrossEntropyDiceLoss, self).__init__()
222-
self.alpha = alpha
223-
224-
def forward(self, y_pred_logits, y_true):
225-
diceloss = MutilDiceLoss(self.alpha)
226-
dice = diceloss(y_pred_logits, y_true)
227-
bceloss = MutilCrossEntropyLoss(self.alpha)
228-
bce = bceloss(y_pred_logits, y_true)
229-
return bce + dice
230-
231-
232215
class LovaszLoss(nn.Module):
233216
"""
234217
mutil LovaszLoss
235218
"""
219+
236220
def __init__(self, per_image=False, ignore=None):
237221
super(LovaszLoss, self).__init__()
238222
self.ignore = ignore

‎model/modelUnet.py

Lines changed: 11 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -5,7 +5,7 @@
55
from .dataset import datasetModelSegwithopencv, datasetModelSegwithnpy
66
from torch.utils.data import DataLoader
77
from .losses import BinaryDiceLoss, BinaryFocalLoss, BinaryCrossEntropyLoss, BinaryCrossEntropyDiceLoss, \
8-
MutilDiceLoss, MutilFocalLoss, MutilCrossEntropyLoss, MutilCrossEntropyDiceLoss
8+
MutilDiceLoss, MutilFocalLoss, MutilCrossEntropyLoss
99
import torch.optim as optim
1010
import numpy as np
1111
from tqdm import tqdm
@@ -709,7 +709,8 @@ def __init__(self, image_depth, image_height, image_width, image_channel, numcla
709709
self.image_channel = image_channel
710710
self.numclass = numclass
711711

712-
self.alpha = [1.] * self.numclass
712+
# self.alpha = [1.] * self.numclass
713+
self.alpha = [1., 5., 1., 5., 3.]
713714
self.gamma = 3
714715

715716
self.use_cuda = use_cuda
@@ -745,8 +746,6 @@ def _loss_function(self, lossname):
745746
return MutilDiceLoss(alpha=self.alpha)
746747
if lossname is 'MutilFocalLoss':
747748
return MutilFocalLoss(alpha=self.alpha, gamma=self.gamma)
748-
if lossname is 'MutilCrossEntropyDiceLoss':
749-
return MutilCrossEntropyDiceLoss(alpha=self.alpha)
750749

751750
def _accuracy_function(self, accuracyname, input, target):
752751
if accuracyname is 'dice':
@@ -772,8 +771,9 @@ def trainprocess(self, trainimage, trainmask, validationimage, validationmask, m
772771
showpixelvalue = showpixelvalue // (self.numclass - 1)
773772
# 1、initialize loss function and optimizer
774773
lossFunc = self._loss_function(self.loss_name)
775-
opt = optim.Adam(self.model.parameters(), lr=lr, weight_decay=1e-4)
776-
lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(opt, 'min', patience=2, verbose=True)
774+
# opt = optim.Adam(self.model.parameters(), lr=lr, weight_decay=1e-4)
775+
opt = optim.Adam(self.model.parameters(), lr=lr)
776+
lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(opt, 'min', patience=10, verbose=True)
777777
# 2、load data train and validation dataset
778778
train_loader = self._dataloder(trainimage, trainmask, True)
779779
val_loader = self._dataloder(validationimage, validationmask)
@@ -793,7 +793,7 @@ def trainprocess(self, trainimage, trainmask, validationimage, validationmask, m
793793
totalValidationLoss = []
794794
totalValiadtionAccu = []
795795
# 4.3、loop over the training set
796-
for batch in train_loader:
796+
for batch in val_loader:
797797
# x should tensor with shape (N,C,D,W,H)
798798
x = batch['image']
799799
# y should tensor with shape (N,C,D,W,H),
@@ -805,6 +805,9 @@ def trainprocess(self, trainimage, trainmask, validationimage, validationmask, m
805805
pred_logit, pred = self.model(x)
806806
loss = lossFunc(pred_logit, y)
807807
accu = self._accuracy_function(self.accuracyname, pred, y)
808+
savepath = model_dir + '/' + str(e + 1) + "_train_EPOCH_"
809+
save_images3d(torch.argmax(pred[0], 0), y[0], showwind, savepath,
810+
pixelvalue=showpixelvalue)
808811
# first, zero out any previously accumulated gradients,
809812
# then perform backpropagation,
810813
# and then update model parameters
@@ -841,7 +844,7 @@ def trainprocess(self, trainimage, trainmask, validationimage, validationmask, m
841844
avgValidationLoss = torch.mean(torch.stack(totalValidationLoss))
842845
avgTrainAccu = torch.mean(torch.stack(totalTrainAccu))
843846
avgValidationAccu = torch.mean(torch.stack(totalValiadtionAccu))
844-
lr_scheduler.step(avgValidationLoss)
847+
# lr_scheduler.step(avgValidationLoss)
845848
# 4.6、update our training history
846849
H["train_loss"].append(avgTrainLoss.cpu().detach().numpy())
847850
H["valdation_loss"].append(avgValidationLoss.cpu().detach().numpy())

‎networks/ResNet2d.py

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -117,13 +117,14 @@ def __init__(self, image_channel, numclass, elu=True):
117117

118118
self.down_tr32 = DownTransition2d(16, 32, 2, elu)
119119
self.down_tr64 = DownTransition2d(32, 64, 3, elu)
120-
self.down_tr128 = DownTransition2d(64, 128, 3, elu, dropout=True)
121-
self.down_tr256 = DownTransition2d(128, 256, 3, elu, dropout=True)
120+
self.down_tr128 = DownTransition2d(64, 128, 3, elu)
121+
self.down_tr256 = DownTransition2d(128, 256, 3, elu)
122122

123123
self.avg = GlobalAveragePooling()
124124

125125
self.fc_layers = nn.Sequential(
126-
nn.Linear(256, 128), nn.ReLU(inplace=True), nn.Dropout(),
126+
nn.Linear(256, 128),
127+
nn.ReLU(inplace=True),
127128
nn.Linear(128, self.numclass))
128129

129130
def forward(self, x):

‎networks/ResNet3d.py

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -117,13 +117,14 @@ def __init__(self, image_channel, numclass, elu=True):
117117

118118
self.down_tr32 = DownTransition3d(16, 32, 2, elu)
119119
self.down_tr64 = DownTransition3d(32, 64, 3, elu)
120-
self.down_tr128 = DownTransition3d(64, 128, 3, elu, dropout=True)
121-
self.down_tr256 = DownTransition3d(128, 256, 3, elu, dropout=True)
120+
self.down_tr128 = DownTransition3d(64, 128, 3, elu)
121+
self.down_tr256 = DownTransition3d(128, 256, 3, elu)
122122

123123
self.avg = GlobalAveragePooling()
124124

125125
self.fc_layers = nn.Sequential(
126-
nn.Linear(256, 128), nn.ReLU(inplace=True), nn.Dropout(),
126+
nn.Linear(256, 128),
127+
nn.ReLU(inplace=True),
127128
nn.Linear(128, self.numclass))
128129

129130
def forward(self, x):

‎networks/Unet3d.py

Lines changed: 5 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -18,15 +18,15 @@ def __init__(self, in_channels, out_channels, init_features=16):
1818
self.pool1 = nn.MaxPool3d(kernel_size=2, stride=2)
1919
self.encoder2 = UNet3d._block(self.features, self.features * 2, name="enc2")
2020
self.pool2 = nn.MaxPool3d(kernel_size=2, stride=2)
21-
self.encoder3 = UNet3d._block(self.features * 2, self.features * 4, name="enc3", dropout=False)
21+
self.encoder3 = UNet3d._block(self.features * 2, self.features * 4, name="enc3")
2222
self.pool3 = nn.MaxPool3d(kernel_size=2, stride=2)
23-
self.encoder4 = UNet3d._block(self.features * 4, self.features * 8, name="enc4", dropout=False)
23+
self.encoder4 = UNet3d._block(self.features * 4, self.features * 8, name="enc4")
2424
self.pool4 = nn.MaxPool3d(kernel_size=2, stride=2)
25-
self.bottleneck = UNet3d._block(self.features * 8, self.features * 16, name="bottleneck", dropout=False)
25+
self.bottleneck = UNet3d._block(self.features * 8, self.features * 16, name="bottleneck")
2626
self.upconv4 = nn.ConvTranspose3d(self.features * 16, self.features * 8, kernel_size=2, stride=2)
27-
self.decoder4 = UNet3d._block((self.features * 8) * 2, self.features * 8, name="dec4", dropout=False)
27+
self.decoder4 = UNet3d._block((self.features * 8) * 2, self.features * 8, name="dec4")
2828
self.upconv3 = nn.ConvTranspose3d(self.features * 8, self.features * 4, kernel_size=2, stride=2)
29-
self.decoder3 = UNet3d._block((self.features * 4) * 2, self.features * 4, name="dec3", dropout=False)
29+
self.decoder3 = UNet3d._block((self.features * 4) * 2, self.features * 4, name="dec3")
3030
self.upconv2 = nn.ConvTranspose3d(self.features * 4, self.features * 2, kernel_size=2, stride=2)
3131
self.decoder2 = UNet3d._block((self.features * 2) * 2, self.features * 2, name="dec2")
3232
self.upconv1 = nn.ConvTranspose3d(self.features * 2, self.features, kernel_size=2, stride=2)
@@ -73,7 +73,6 @@ def _block(in_channels, features, name, dropout=False):
7373
bias=False, ),),
7474
(name + "norm1", nn.BatchNorm3d(num_features=features)),
7575
(name + "relu1", nn.ReLU(inplace=True)),
76-
(name + "dropout1", nn.Dropout3d()),
7776
(name + "conv2", nn.Conv3d(
7877
in_channels=features,
7978
out_channels=features,

0 commit comments

Comments
 (0)