Binarized Mamba-Transformer for Lightweight Quad Bayer HybridEVS Demosaicing (CVPR 2025)

This repository contains the official PyTorch implementation for the paper: "Binarized Mamba-Transformer for Lightweight Quad Bayer HybridEVS Demosaicing" accepted at CVPR 2025.

Our work, BMTNet, introduces a lightweight Mamba-based binary neural network for efficient and high-performance demosaicing of HybridEVS RAW images. We propose a hybrid Binarized Mamba-Transformer architecture to effectively capture both global and local dependencies.

Installation

We recommend using Miniconda for environment management.

Step 1. Clone the repository

git clone https://github.com/Clausy9/BMTNet.git
cd BMTNet

Step 2. Create and activate the conda environment

conda create --name bmtnet python=3.8 -y
conda activate bmtnet

Step 3. Install dependencies

Install PyTorch according to the official instructions for your specific CUDA version. For example:

# Example for CUDA 11.8
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118```

Then, install the remaining required packages:

```bash
pip install -r requirements.txt

The requirements.txt file should contain:

argparse
tqdm
scikit-image
opencv-python
einops
timm
thop
random-fourier-features-pytorch
mamba-ssm==2.2.1

Prerequisites

• Linux
• NVIDIA GPU
• PyTorch 1.12+
• CUDA 11.6+

Testing

Step 1. Download pretrained models

Download the pretrained models from [Google Drive] and place them in the ./weights directory.

Step 2. Prepare your data

Place your input images in the ./input directory.

Step 3. Run the testing script

To run the testing script, use the following command. The output will be saved in the ./output directory.

python test.py --input_path ./input --weights_path ./weights/bmtnet_model_latest.pth --save_path ./output

You can adjust the input and output paths as needed.