LGMSNet

LGMSNet is a lightweight framework for 2D and 3D medical image segmentation, implemented in PyTorch. It provides a complete pipeline for training, validation, and testing, with support for multiple datasets.

LGMSNet: Thinning a medical image segmentation model via dual-level multiscale fusion accepted by ECAI 2025 oral.

Project Structure

SegmsNet/  
├── checkpoint/               # Saved model weights  
├── data/                     # Dataset directory  
├── dataloader/               # Data loader modules  
├── network/                  # 2D segmentation models  
├── network_3d/               # 3D segmentation models  
├── src/                      # Source code  
├── utils/                    # Utility functions and scripts  
├── datamodule_my.py          # Custom data module  
├── environment.yaml          # Environment configuration file  
├── main_kvasir.py            # Main script for Kvasir dataset  
├── main.py                   # Main script for 2D segmentation  
├── main3d.py                 # Main script for 3D segmentation  
├── train.sh                  # Training script  
├── trainer.py                # Trainer module  
└── README.md                 # Project documentation  

Environment Setup

1. Install Dependencies

Ensure conda is installed, then create the environment using:

conda env create -f environment.yaml  
conda activate uxnet3d  

Dataset Preparation

Place datasets in the data/ directory. Supported datasets include:

• Kvasir
• BUSI
• TNSCUI
• ISIC18
• BTCV
• KiTS23

Data can be found in U-Bench

Model Training

Run the train.sh script, which includes training configurations for all supported datasets.

bash train.sh