STDDNet

Official PyTorch implementation of "STDDNet: Harnessing Mamba for Video Polyp Segmentation via Spatial-aligned Temporal Modeling and Discriminative Dynamic Representation Learning".

Getting Start

Installation

1. Envs: python=3.10.13 and CUDA=11.8

conda create -n stddnet python=3.10.13
conda activate stddnet
pip install torch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 --index-url https://download.pytorch.org/whl/cu118
pip install numpy==1.26.3 timm==0.4.12 einops==0.7.0 packaging==23.2 tqdm
pip install -U scikit-learn
conda install -c conda-forge opencv

2. Requirements (if necessary)

Please refer to Vim to download vim_requirements.txt

pip install -r your_path/vim_requirements.txt

3. Install Causal_conv1d and Mamba_ssm (Two options)

A. Following the setup of Vim

  pip install -e causal_conv1d>=1.1.0
  pip install -e mamba-1p1p1

B. Download and Install causal_conv1d-1.4.0+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl and mamba_ssm-1.1.1+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl

Downloads:
https://github.com/Dao-AILab/causal-conv1d/releases?page=2
https://github.com/state-spaces/mamba/releases?page=3

Installation:
pip install your_path/causal_conv1d-1.4.0+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
pip install your_path/mamba_ssm-1.1.1+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl

Then,

a. replace the compiled file mamba_simple.py (at /your_home_path/anaconda3/envs/stddnet/lib/python3.10/site-packages/mamba_ssm/modules/)

b. replace the compiled file selective_scan_interface.py (at /your_home_path/anaconda3/envs/stddnet/lib/python3.10/site-packages/mamba_ssm/ops/)

Training and Testing

//Train:
python /your_project_path/scripts/my_train.py
//Test:
python /your_project_path/scripts/my_test.py

Logs and Weights

We provide the relevant logs and ckpts (trained on SUN-SEG dataset) based on two different backbones: Res2Net-50 and PVTv2-B2

Acknowledgement

Our work builds upon the excellent foundational research of PNS+ and Vim. We thank the authors for their awesome works and publicly available codes.

Citation

if you find our work useful, please cite:

@inproceedings{chen2025stddnet,
  title={STDDNet: Harnessing Mamba for Video Polyp Segmentation via Spatial-aligned Temporal Modeling and Discriminative Dynamic Representation Learning},
  author={Chen, Guilian and Wu, Huisi and Qin, Jing},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={21364--21373},
  year={2025}
}