MobileViM: A Light-weight and Dimension-independent Vision Mamba for 3D Medical Image Analysis

Welcome to the official implementation of ``MobileViM: A Light-weight and Dimension-independent Vision Mamba for 3D Medical Image Analysis''. This repository provides a comprehensive toolkit optimized for deep learning and computer vision applications, particularly focusing on 3D semantic segmentation. It includes functionalities for visualizing training progress, logging activities, and computing standard performance metrics.

Architecture overview.

Dimension-independent mechanism.

Dual-direction Mamba.

The manuscript is currently under review at a peer-reviewed journal. The full code is now publicly accessible.

Setup

For installation instructions of MobileViM, please consult the comprehensive guide in INSTALL.md.

Benchmarking and Evaluation

Refer to DATA.md for detailed instructions on data preparation for benchmarking and model training.

To begin training and evaluation, follow the configuration details in the ours.sh script.

Ablation studies

For specifics on ablation studies and further experiments, please see the scripts in ablation.sh.

Citing Our Work

If our implementation aids your research, please acknowledge it by citing our paper:

@misc{dai2025mobilevim,
  title={MobileViM: A Light-weight and Dimension-independent Vision Mamba for 3D Medical Image Analysis},
  author={Dai, Wei, Hui, Chen, and Liu, Jun},
  year={2025},
  eprint={2502.13524},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2502.13524}, 

}