Alex8211101/CLD-Semi

MICCAI 2022 (Provisionally Accepted): Calibrating Label Distribution for Class-Imbalanced Barely-Supervised Knee Segmentation

CLD-Semi

Yiqun Lin, Huifeng Yao, Zezhong Li, Guoyan Zheng, Xiaomeng Li, "Calibrating Label Distribution for Class-Imbalanced Barely-Supervised Knee Segmentation", MICCAI 2022 (Provisionally Accepted). [paper]

0. Citation

@misc{cld2022lin,
  doi = {10.48550/ARXIV.2205.03644},
  url = {https://arxiv.org/abs/2205.03644},
  author = {Lin, Yiqun and Yao, Huifeng and Li, Zezhong and Zheng, Guoyan and Li, Xiaomeng},
  keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {Calibrating Label Distribution for Class-Imbalanced Barely-Supervised Knee Segmentation},
  publisher = {arXiv},
  year = {2022}
}

1. Environment

This code has been tested with Python 3.6, PyTorch 1.8, torchvision 0.9.0, and CUDA 11.1 on Ubuntu 20.04.

2. Data Preparation

The MR imaging scans are available at https://oai.nih.gov/. Run the function process_npy in ./code/data/preprocess.py to convert .nii.gz files into .npy for faster loading. To generate the labeled/unlabeled splits, run the function process_split_semi or use our pre-split files in ./knee_data/splits/*.txt. After preprocessing, the ./knee_data/ folder should be organized as follows:

./knee_data/
├── imagesTr
│   ├── <id>_0000.nii.gz
├── labelsTr
│   ├── <id>.nii.gz
├── imagesTs
│   ├── <id>_0000.nii.gz
├── labelsTs
│   ├── <id>.nii.gz
├── npy
│   ├── <id>_image.npy
│   ├── <id>_label.npy
├── splits
│   ├── labeled.txt
│   ├── unlabeled.txt
│   ├── train.txt
│   ├── eval.txt
│   ├── test.txt

3. Training

Run the following commands for training.

mkdir -p ./logs/__nohup

bash py_run.sh code/train_cld.py --exp cld -g 0

4. Testing

Run the following commands for testing.

bash py_run.sh code/test.py --exp cld -ep 280 --cps A -g 0
python code/evaluate.py -p ./logs/cld/predictions/ep_280/

Model	Avg.	DF	FC	Ti	TC	Link
CLD-Semi	87.2	93.8	83.7	92.8	78.6	ep_280.pth

License

This repository is released under MIT License (see LICENSE file for details).