Linfeng-Tang/Image-Fusion

What it solves

This repository serves as a comprehensive collection of research papers, datasets, and implementations for image fusion—the process of combining multiple source images (such as infrared and visible light, multi-exposure, or multi-focus images) into a single composite image that retains the most salient features from each.

How it works

The project organizes a vast library of image fusion methods categorized by their primary purpose and technical approach:

• Multi-Modal Fusion: Focuses on combining different sensor types, such as infrared (thermal) and visible light images, often used for target detection and surveillance.
• Digital Photography Fusion: Includes multi-exposure and multi-focus fusion to improve dynamic range or depth of field.
• Remote Sensing Fusion: Covers pansharpening to enhance resolution in satellite imagery.
• Technical Frameworks: The collection tracks methods using various architectures including Convolutional Neural Networks (CNNs), Transformers, Generative Adversarial Networks (GANs), Diffusion Models, and Vision-Language Models (VLMs).

Who it’s for

• Computer Vision Researchers: Those studying low-level vision, image processing, and multi-modal data integration.
• AI Engineers: Developers looking for state-of-the-art (SOTA) implementations of image fusion algorithms for robotics, surveillance, or medical imaging.
• Students: Learners seeking a structured survey of the field from early autoencoders to modern diffusion-based controllable fusion.

Highlights

• Extensive Taxonomy: Categorizes methods into visual-oriented, semantic-driven, joint registration-fusion, and degradation-robust fusion.
• LVM Integration: Tracks the latest trends in using Large Vision Models and language-driven prompts for controllable image fusion.
• Comprehensive Resources: Provides direct links to papers and code for dozens of high-impact publications in venues like CVPR, TPAMI, and NeurIPS.