IRSRMamba

Official PyTorch implementation of the paper IRSRMamba: Infrared Image Super-Resolution via Mamba-based Wavelet Transform Feature Modulation Model.

• 🥱Please note that this code repository will no longer be updated or maintained.
• 🚩Accepted by IEEE TGRS 💡 If you found this helpful, please consider citing our work! Thank you!
• 🍰Stronger Mamba-based model: GPSMamba has been released. 💡 Please refer to: GPSMamba.

Introduction

Infrared image super-resolution demands long-range dependency modeling and multi-scale feature extraction to address challenges such as homogeneous backgrounds, weak edges, and sparse textures. While Mamba-based state-space models (SSMs) excel in global dependency modeling with linear complexity, their block-wise processing disrupts spatial consistency, limiting their effectiveness for IR image reconstruction. We propose IRSRMamba, a novel framework integrating wavelet transform feature modulation for multi-scale adaptation and an SSMs-based semantic consistency loss to restore fragmented contextual information. This design enhances global-local feature fusion, structural coherence, and fine-detail preservation while mitigating block-induced artifacts. Experiments on benchmark datasets demonstrate that IRSRMamba outperforms state-of-the-art methods in PSNR, SSIM, and perceptual quality. This work establishes Mamba-based architectures as a promising direction for high-fidelity IR image enhancement.

Approach overview

Please check here.

Requirements

• Python 3.8, PyTorch >= 1.11
• BasicSR 1.4.2
• Platforms: Ubuntu 18.04, cuda-11
• einops pip install einops
• fvcore pip install -U fvcore

Installation

Clone the repo

git clone https://github.com/yongsongH/IRSRMamba.git

Install dependent packages

cd IRSRMamba

pip install -r install.txt

Install BasicSR

python setup.py develop

You can refer to MambaIR for mamab-ssm installation

Dataset Acquisition

The dataset used in this project can be obtained from the following link: JinyuanLiu-CV/TarDAL Please note that data ownership remains with the original dataset authors. We are grateful for their work and contributions.

Dataset Usage

For this project, we selected the first 265 samples from the dataset, in sequential order, to form the training dataset for model training.

HR-LR Pair Generation

The process for generating High-Resolution (HR) and Low-Resolution (LR) pairs from the dataset can be referenced in the following script: bic_dataset.py Please refer to the script for detailed steps on how the HR-LR pairs were processed.

Data Ownership

Please ensure you comply with the terms of use and licensing agreements provided by the original dataset authors when using this dataset.

Model

Pre-trained models can be downloaded from this link.

Evaluation

please check the log file for more information about the settings.

Training

• Run the following commands for training:

python basicsr/train.py -opt options/train/train_IRSRMamba_final_x2.yml

python basicsr/train.py -opt options/train/train_IRSRMamba_final_x4.yml

Testing

• Download the pretrained models.
• Download the testing dataset.
• Run the following commands:

  python basicsr/test.py -opt options/test/test_IRSRMamba_x4.yml

  python basicsr/test.py -opt options/test/test_IRSRMamba_x2.yml

Contact

For some unpleasant reasons, I will be unable to continue maintaining this code repository. Please feel free to develop a better version according to your needs, in compliance with the license agreement. Some of the resolved issues might be helpful. I sincerely wish you all the best in your research and career.

Citation

@ARTICLE{11059944,
  author={Huang, Yongsong and Miyazaki, Tomo and Liu, Xiaofeng and Omachi, Shinichiro},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={IRSRMamba: Infrared Image Super-Resolution via Mamba-Based Wavelet Transform Feature Modulation Model}, 
  year={2025},
  volume={63},
  number={},
  pages={1-16},
  doi={10.1109/TGRS.2025.3584385}}

Acknowledgement

I would like to express my sincere gratitude to all the authors whose work has contributed to this project. Their insights and contributions have been invaluable. I am also deeply thankful for the positive feedback and constructive assistance received from the editors and reviewers. Their expertise and thoughtful suggestions have significantly improved the quality and clarity of this work.

---

Legal Action and Patent Clause

This project is licensed under the Apache License, Version 2.0. A copy of the Apache License, Version 2.0, is included in the LICENSE file in this repository. You may also obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0.

Any violation of the terms of the Apache License, Version 2.0, may result in legal action and liability for damages.