MambaIC: State Space Models for High-Performance Learned Image Compression (CVPR 2025)

This repo is the official implementation of CVPR 2025 paper: MambaIC: State Space Models for High-Performance Learned Image Compression

MambaIC: State Space Models for High-Performance Learned Image Compression

Fanhu Zeng, Hao Tang, Yihua Shao, Siyu Chen, Ling Shao, Yan Wang

Key words: Learned image compression, State space model, Context model.

✨ News

• [2025.03.18] We release camera-ready submission on arxiv. 🍰
• [2025.03.15] Github repo is available. 🍬
• [2025.02.27] MambaIC has been accepted by CVPR 2025! 🎉

📖 Abstract

A high-performance image compression algorithm is crucial for real-time information transmission across numerous fields. Despite rapid progress in image compression, computational inefficiency and poor redundancy modeling still pose significant bottlenecks, limiting practical applications. Inspired by the effectiveness of state space models (SSMs) in capturing long-range dependencies, we leverage SSMs to address computational inefficiency in existing methods and improve image compression from multiple perspectives. In this paper, we integrate the advantages of SSMs for better efficiency-performance trade-off and propose an enhanced image compression approach through refined context modeling, which we term MambaIC. Specifically, we explore context modeling to adaptively refine the representation of hidden states. Additionally, we introduce window-based local attention into channel-spatial entropy modeling to reduce potential spatial redundancy during compression, thereby increasing efficiency. Comprehensive qualitative and quantitative results validate the effectiveness and efficiency of our approach, particularly for high-resolution image compression.

🏠 Architecture

Overall structure of MambaIC.

Illustration of the proposed SSM-based context entropy model with window-based local attention.

Environmental Setup

Install CompressAI

git clone https://github.com/InterDigitalInc/CompressAI compressai
cd compressai
pip install -U pip && pip install -e .

Install Vmamba

git clone https://github.com/MzeroMiko/VMamba.git
cd VMamba
cd kernels/selective_scan && pip install .

Dataset

MambaIC
|-- dataset
    |-- flickr30k
        |-- train_1.jpg
        |-- train_2.jpg
        ...
    |-- Kodak
        |-- kodak_1.jpg
        |-- kodak_2.jpg
        ...
    |-- CLIC
        |-- CLIC_1.jpg
        |-- CLIC_2.jpg
        ... 
    |-- Tecnick
        |-- Tecnick_1.jpg
        |-- Tecnick_2.jpg
        ...

Training

Evaluation

Experimental Results

Quantitative RD Results on Kodak (left: PSNR, Right: MS-SSIM).

📘 Citation

If you find this work useful, consider giving this repository a star ⭐ and citing 📑 our paper as follows:

@article{zeng2025mambaic,
  title={MambaIC: State Space Models for High-Performance Learned Image Compression},
  author={Zeng, Fanhu and Tang, Hao and Shao, Yihua and Chen, Siyu and Shao, Ling and Wang, Yan},
  journal={arXiv preprint arXiv:2503.12461},
  year={2025}
}

Note

This is not the exact original code and is a re-implementation of our CVPR 2025 paper. But the core code and experimental results are almost the same, with slight difference.

Acknowledgememnt

The code is based on CompressAI, Mamba, Vmamba, MambaVision, ELIC, MambaVC. Thanks for these great works and open sourcing!

If you find them helpful, please consider citing them as well.