Awesome

MagicNet: Semi-Supervised Multi-Organ Segmentation via Magic-Cube Partition and Recovery (CVPR 2023)

Introduction

MagicNet: Semi-Supervised Multi-Organ Segmentation via Magic-Cube Partition and Recovery, <br/> Duowen Chen, Yunhao Bai, Wei Shen, Qingli Li, Lequan Yu and Yan Wang. <br/> In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023 <br/> [arXiv][bibtex][supp]

News

• [2024.11.06] We have released our checkpoints and training logs on model checkpoint for 30% and 40% BTCV setting. And if you have any questions, welcome contact me at 'duowen_chen@hotmail.com'~~~
• [2023.12.15] We have updated 'cube_losses.py'.
• [2023.10.25] We have uploaded the data-splitting file 'btcv.txt' for BTCV dataset to help you reproduce/follow our work^_^!
• [2023.07.01] We have updated the preprocessed data!
• [2023.06.25] Our codes are released!
• [2023.03.16] Repo created. Paper and code will come soon.

Installation

• PyTorch 1.12.0
• CUDA 11.3
• Python 3.8.13

Usage

Dataset and Pre-processing

The datasets used in our paper are MACT dataset and BTCV dataset. You can download directly our preprocessed data from [baidu netdisk](https://pan.baidu.com/s/1OVbDXzE_XaTtFGeILQtRyQ (password: 638u).

Training Steps

1. Clone the repo and create data path:

git clone https://github.com/DeepMed-Lab-ECNU/MagicNet.git
cd MagicNet
mkdir data # create data path

2. Put the preprocessed data in ./data/MACT_h5 for MACT dataset. (./data/btcv_h5 for BTCV dataset) and then cd code
3. We train our model on one single NVIDIA 3090 GPU for each dataset.

To produce the claimed results for MACT dataset:

# For 10% labeled data,
CUDA_VISIBLE_DEVICES=0 python train_main_mact.py --labelnum=7

# For 20% labeled data, 
CUDA_VISIBLE_DEVICES=0 python train_main_mact.py --labelnum=13

To produce the claimed results for BTCV dataset:

# For 30% labeled data,
CUDA_VISIBLE_DEVICES=0 python train_main_btcv.py --labelnum=5

# For 40% labeled data, 
CUDA_VISIBLE_DEVICES=0 python train_main_btcv.py --labelnum=7

Citation

If this code is useful for your research, please consider giving star to our repository and citing our work:

@InProceedings{Chen_2023_CVPR, 
	author = {Chen, Duowen and Bai, Yunhao and Shen, Wei and Li, Qingli and Yu, Lequan and Wang, Yan}, 
	title = {MagicNet: Semi-Supervised Multi-Organ Segmentation via Magic-Cube Partition and Recovery}, 
	booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, 
	month = {June}, 
	year = {2023}, 
	pages = {23869-23878} 
}

Questions

If you have any questions, welcome contact me at 'duowen_chen@hotmail.com'