GRDBIS: Graph Relation Distillation for Efficient Biomedical Instance Segmentation

Under Review Arxiv

This repo contains the code of our paper submitted, which is an extension version of our work "Efficient Biomedical Instance Segmentation via Knowledge Distillation" published in MICCAI-22.

Installaion

This code was implemented with Pytorch 1.0.1 (later versions may work), CUDA 9.0, Python 3.7.4.

If you have a Docker environment, we strongly recommend you to pull our image as follows:

docker pull registry.cn-hangzhou.aliyuncs.com/em_seg/v54_higra

Training

2D dataset, take the network MobileNetV2 on the C_elegans dataset as an example

 python trainKD_Celegans.py -c=Celegans_ResUNet_MobileNetV2

3D dataset, take the network MALA-tiny on the AC3/4 dataset as an example

 python trainKD_EM.py -c=AC34_MALA_MALA-tiny

Testing

2D dataset, take the network MobileNetV2 on the C_elegans dataset as an example

 python inference_Celegans.py -mn=ResUNet_MobileNetV2_KD

3D dataset, take the network MALA-tiny on the AC3/4 dataset as an example

 python inference_EM.py -mn=MALA-tiny_KD

Pretrained Model

The related pretrained models are available, please refer to the testing command for evaluation.

Workflow

Contact

If you have any problem with the released code, please contact me by email ([email protected]).