TriMix: A General Framework for Medical Image Segmentation from Limited Supervision

Introduction

This repository provides PyTorch implementation of our ACCV2022 paper 'TriMix: A General Framework for Medical Image Segmentation from Limited Supervision'.

Usage

[1]. Prepare the dataset

Please follow Luo et al. to prepare the ACDC dataset and put it in ./TriMix/semi_supervised/2D/dataset and ./TriMix/scribble_supervised/ACDC/dataset

Please follow CycleMix to prepare the MSCMRSeg dataset and put it in ./TriMix/scribble_supervised/MSCMR/dataset

Please follow UA-MT to prepare the LA dataset and put it in ./TriMix/semi_supervised/3D/dataset

[2]. Train and test the model:

python train_trimix.py
python test_trimix.py

Acknowledgement

Benchmarks on ACDC, MSCMRSeg, and LA datasets in this repository borrow part of codes from Luo et al., CycleMix, and UA-MT implementations. We also appreciate other public codebases cited in our paper.

Note

Contact: Zhou Zheng (zzheng@mori.m.is.nagoya-u.ac.jp)