Source code for Few-Shot Transfer Learning for Hereditary Retinal Diseases Recognition(early accepted in MICCAI 2021).

@inproceedings{mai2021few,
  title={Few-Shot Transfer Learning for Hereditary Retinal Diseases Recognition},
  author={Mai, Siwei and Li, Qian and Zhao, Qi and Gao, Mingchen},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={97--107},
  year={2021},
  organization={Springer}
}

In our experiments, we used ResNet50 as the teacher model and ResNet18 as the student model by default. We also integrated many other commonly used models as additional choices in the models folder, please modify the individual training files if you select them.

For confidentiality reasons, we do not provide the target dataset in the experiment, but the BOE dataset and Cell dataset are publicly available for download. For training, the dataset directory needs to be modified in the corresponding dataloader files.

python3 preprocess.py --path <path to dataset>
*OpenBLAS is optional to install for accelerating the calculation.

python3 train_projector.py -T 50 -a -5.0 --info "trial 1"

python3 transfer2teacher.py --methods (ALL/HL/FE) --path_t *path to the .pth file trained by projector*"

python3 train_student.py --path_t *path to the .pth file trained by teacher*"

Still in the process of improvement, if you find any problems, or have any questions, please feel free to contact me ([email protected])