unet liver

Unet network for liver CT image segmentation

data preparation

structure of project

  --project
  	main.py
  	 --data
   		--train
   		--val

data and trained weight link: https://pan.baidu.com/s/1dgGnsfoSmL1lbOUwyItp6w code: 17yr

all dataset you can access from: https://competitions.codalab.org/competitions/15595

training

python main.py train

testing

load the last saved weight

python main.py test --ckpt=weights_19.pth

数据准备

项目文件分布如下

  --project
  	main.py
  	 --data
   		--train
   		--val

数据和权重可以使用百度云下载 链接:

链接: https://pan.baidu.com/s/1dgGnsfoSmL1lbOUwyItp6w 提取码: 17yr

全部数据集: https://competitions.codalab.org/competitions/15595

模型训练

python main.py train

测试模型训练

加载权重，默认保存最后一个权重

python main.py test --ckpt=weights_19.pth

多类别

修改2个地方即可：unet最后一层的通道数设置为类别数；损失函数使用CrossEntropyLoss

bath_size,img_size,num_classes=2,3,4
#model = Unet(3, num_classes)
criterion = nn.CrossEntropyLoss()
#assume the pred is the output of the model
pred=torch.rand(bath_size,num_classes,img_size,img_size)
target=torch.randint(num_classes,(bath_size,img_size,img_size))
loss=criterion(pred,target)

Demo