MaeFuse Checkpoint

English

This repository hosts the Hugging Face checkpoint for MaeFuse, the official implementation of “MaeFuse: Transferring Omni Features With Pretrained Masked Autoencoders for Infrared and Visible Image Fusion via Guided Training.”

MaeFuse is designed for infrared-visible image fusion (IVIF). This repository is intended to provide a convenient way to download and use the released checkpoint on Hugging Face.

Official Resources

• Official GitHub repository: Henry-Lee-real/MaeFuse
• Original project page: https://github.com/Henry-Lee-real/MaeFuse

Overview

MaeFuse proposes using a pretrained masked autoencoder as a high-level semantic encoder for image fusion. According to the official project description, the method aims to avoid unnecessary downstream-task complexity while preserving strong semantic guidance for infrared-visible fusion.

This Hugging Face repository mainly provides the checkpoint for easier access and deployment. For the full training code, testing scripts, and implementation details, please refer to the official GitHub repository.

Environment

The official project recommends:

• Python 3.10
• Install dependencies with:

pip install -r requirements.txt

Inference

According to the official repository, the dataset directory passed to --address should contain two subdirectories:

• vi/ for visible images
• ir/ for infrared images

Example inference command:

python test_fusion.py --checkpoint path_to_weight --address path_to_dataset --output path_to_output

Argument description:

• --checkpoint: path to the model checkpoint file
• --address: path to the dataset directory
• --output: directory for saving fused results

Notes

• This repository mainly hosts the checkpoint for convenient access on Hugging Face.
• For training details, pretraining guidance, and the latest official updates, please refer to the original GitHub repository.
• You may further edit this README to add exact checkpoint filenames, example images, evaluation results, or loading scripts specific to this release.

Citation

@ARTICLE{10893688,
  author={Li, Jiayang and Jiang, Junjun and Liang, Pengwei and Ma, Jiayi and Nie, Liqiang},
  journal={IEEE Transactions on Image Processing},
  title={MaeFuse: Transferring Omni Features With Pretrained Masked Autoencoders for Infrared and Visible Image Fusion via Guided Training},
  year={2025},
  volume={34},
  pages={1340-1353},
  doi={10.1109/TIP.2025.3541562}
}

---

中文

本仓库提供 MaeFuse 的 Hugging Face 权重文件，方便用户直接下载与使用。MaeFuse 对应论文为 《MaeFuse: Transferring Omni Features With Pretrained Masked Autoencoders for Infrared and Visible Image Fusion via Guided Training》，任务方向为红外-可见光图像融合。

官方资源

• 官方 GitHub 仓库：Henry-Lee-real/MaeFuse
• 项目地址：https://github.com/Henry-Lee-real/MaeFuse

方法简介

根据官方项目说明，MaeFuse 的核心思路是使用预训练的 Masked Autoencoder 作为具备高层语义信息的编码器，用于红外与可见光图像融合，从而避免过于复杂的下游任务设计，同时保留较强的语义表达能力。

本 Hugging Face 仓库主要用于提供 checkpoint 下载与部署支持。若需查看完整训练代码、测试脚本和实现细节，请参考官方 GitHub 仓库。

环境要求

官方仓库建议：

• Python 3.10
• 使用以下命令安装依赖：

pip install -r requirements.txt

推理方式

根据官方仓库说明，--address 指向的数据集目录下应包含两个子目录：

• vi/：可见光图像
• ir/：红外图像

示例命令：

python test_fusion.py --checkpoint path_to_weight --address path_to_dataset --output path_to_output

参数说明：

• --checkpoint：模型权重路径
• --address：测试数据路径
• --output：融合结果保存目录

说明

• 本仓库主要用于在 Hugging Face 上提供权重下载。
• 训练细节、预训练流程说明以及最新更新，请以官方 GitHub 仓库为准。
• 你也可以继续补充当前 release 的具体权重名、示例图片、测试结果或加载脚本。

引用

@ARTICLE{10893688,
  author={Li, Jiayang and Jiang, Junjun and Liang, Pengwei and Ma, Jiayi and Nie, Liqiang},
  journal={IEEE Transactions on Image Processing},
  title={MaeFuse: Transferring Omni Features With Pretrained Masked Autoencoders for Infrared and Visible Image Fusion via Guided Training},
  year={2025},
  volume={34},
  pages={1340-1353},
  doi={10.1109/TIP.2025.3541562}
}