Unet network for liver CT image segmentation
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
python main.py train
load the last saved weight
python main.py test --ckpt=weights_19.pth
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
python main.py train
加载权重，默认保存最后一个权重
python main.py test --ckpt=weights_19.pth
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)