- (四:2020.07.28)nnUNet最舒服的训练教程(让我的奶奶也会用nnUNet(上))(21.04.20更新)
https://blog.csdn.net/weixin_42061636/article/details/107623757
- 保姆级教程:nnUnet在2维图像的训练和测试
https://blog.csdn.net/minervazhaojie/article/details/112061000
- (十七:2020.09.10)nnUNet最全问题收录(9.10更新)
https://blog.csdn.net/weixin_42061636/article/details/108520695
- 安装hiddenlayer(用来生成什么网络拓扑图?管他呢,装吧)
pip install --upgrade git+https://github.com/nanohanno/hiddenlayer.git@bugfix/get_trace_graph#egg=hiddenlayer
git clone https://github.com/MIC-DKFZ/nnUNet.git
cd nnUNet
pip install -e .
export nnUNet_raw_data_base="/home/qiao/nnUNetFrame/DATASET/nnUNet_raw"
export nnUNet_preprocessed="/home/qiao/nnUNetFrame/DATASET/nnUNet_preprocessed"
export RESULTS_FOLDER="/home/qiao/nnUNetFrame/DATASET/nnUNet_trained_models"
nnUNetFrame
├──DATASET
├── dataset.json
├── imagesTr
│ ├── Case100_10_0000.nii.gz
│ ├── Case100_11_0000.nii.gz
│ ├── ...
├── imagesTs
│ ├── Case161_10_0000.nii.gz
│ ├── Case161_11_0000.nii.gz
│ ├── ...
└── labelsTr
- 下载开源数据集(建议04,最小),我使用自己的,因此需要照着改
https://drive.google.com/drive/folders/1HqEgzS8BV2c7xYNrZdEAnrHk7osJJ--2
- 修改labels,numTraining,numTest
- Task后面的id,必须是三位数
- 在dataset.json中修改:
"training":[{"image":"./imagesTr/Case100_1.nii.gz","label":"./labelsTr/Case100_1.nii.gz"},...]
"test":["./imagesTs/Case161_1.nii.gz",...]
- 注意:
- 对于文件:imagesTr和imagesTs需要加上
_0000.nii.gz
,labelsTr不需要加,但是要和imagesTr名字对应
- 对于dataset.json:都不需要加
_0000
nnUNet_raw_data_base/nnUNet_raw_data/Task058_Spine
├── dataset.json
├── imagesTr
│ ├── Case100_10_0000.nii.gz
│ ├── Case100_11_0000.nii.gz
│ ├── ...
├── imagesTs
│ ├── Case161_10_0000.nii.gz
│ ├── Case161_11_0000.nii.gz
│ ├── ...
└── labelsTr
├── Case100_1.nii.gz
├── Case100_10.nii.gz
├── ...
nnUNet_plan_and_preprocess -t 058 -pl3d None --verify_dataset_integrity
nnUNet_train 2d nnUNetTrainerV2 058 all
nohup nnUNet_train 2d nnUNetTrainerV2 058 all > runlog.txt 2>&1 &