This repository serves as a comprehensive survey and benchmark for diffusion model-based image editing techniques. It categorizes and analyzes a wide range of methods, offering a structured overview for researchers and practitioners in the field of generative AI and computer vision. The project also introduces EditEval, a benchmark for evaluating text-guided image editing, and its associated LMM Score metric.

How It Works

The survey categorizes diffusion model-based image editing methods into three main strategies: Training-Based, Testing-Time Finetuning, and Training and Finetuning Free. Each category is further broken down by specific techniques like domain-specific editing, reference guidance, instructional editing, latent variable optimization, and attention modification. This structured approach allows for a systematic understanding of the diverse methodologies employed in diffusion model image editing.

Quick Start & Requirements

This repository is primarily a survey and benchmark, not a runnable codebase for a specific editing method. The README provides links to download the EditEval_v1 and EditEval_v2 datasets for evaluation purposes.

Highlighted Details

• Comprehensive survey paper accepted by TPAMI 2025.
• Introduces EditEval_v1 and EditEval_v2 benchmarks for evaluating image editing algorithms.
• Proposes the LMM Score metric for evaluating text-guided image editing performance.
• Includes a leaderboard comparing methods using the LMM Score.

Maintenance & Community

The project is actively tracking research and welcomes contributions. Updates are posted regularly, including the release of EditEval_v2 and the LMM Score template.

Licensing & Compatibility

The repository itself does not specify a license. The survey paper is cited with standard academic citation practices.

Limitations & Caveats

This repository is a survey and benchmark collection, not a unified framework or library for performing image editing. Users will need to refer to individual papers for implementation details and code.