PRE-Mamba: A 4D State Space Model for Ultra-High-Frequent Event Camera Deraining

ICCV 2025

PRE-Mamba

Demo video

Overview

• Environment
• Dataset preparation
• Run
• Citation
• Acknowledgement

Environment

# pointcept with CUDA=11.6
conda create -n pointcept python=3.8 -y
conda activate pointcept
conda install ninja -y
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge
conda install h5py pyyaml -c anaconda -y
conda install sharedarray tensorboard tensorboardx yapf addict einops scipy plyfile termcolor timm -c conda-forge -y
conda install pytorch-cluster pytorch-scatter pytorch-sparse -c pyg -y

pip install torch-geometric
pip install spconv-cu116
pip install open3d

cd libs/pointops
python setup.py install
cd ../../

# mamba install
cd libs/
git clone https://github.com/Dao-AILab/causal-conv1d.git
cd causal-conv1d
git checkout v1.1.3 
CAUSAL_CONV1D_FORCE_BUILD=TRUE pip install .
cd ..
git clone https://github.com/state-spaces/mamba.git
cd mamba
git checkout v1.1.4 
MAMBA_FORCE_BUILD=TRUE pip install .

Dataset preparation

Our dataset EventRain-27K is now available on Hugging Face.

Run

Training

export CUDA_VISIBLE_DEVICES=${CUDA_VISIBLE_DEVICES}
bash scripts/train.sh -g ${NUM_GPU} -d ${DATASET_NAME} -c ${CONFIG_NAME} -n ${EXP_NAME}

For example:

export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5
bash scripts/train.sh -g 6 -d event_rain -c PRE_Mamba -n SYTHETIC

Testing

bash scripts/test.sh -g ${NUM_GPU} -d ${DATASET_NAME} -c ${CONFIG_NAME} -n ${EXP_NAME} -w ${CHECKPOINT_NAME}

For example:

export CUDA_VISIBLE_DEVICES=0
# weight path: ./exp/event_rain/SYTHETIC/model/model_best.pth
bash scripts/test.sh -g 1 -d event_rain -c PRE_Mamba -n SYTHETIC -w model_best

Citation

If you find this repository helpful to your research, please consider citing the following:

@InProceedings{Ruan_2025_ICCV,
    author    = {Ruan, Ciyu and Guo, Ruishan and Gong, Zihang and Xu, Jingao and Yang, Wenhan and Chen, Xinlei},
    title     = {PRE-Mamba: A 4D State Space Model for Ultra-High-Frequent Event Camera Deraining},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2025},
    pages     = {9169-9180}
}

Acknowledgement

The code framework of this project is based on pointcept and MambaMOS, thanks to their excellent work.