ImgEdit: A Unified Image Editing Dataset and Benchmark
Yang Ye, Xianyi He, Zongjian Li, Bin Lin, Shenghai Yuan, Zhiyuan Yan, Bohan Hou, Li Yuan
2025-05-28
Summary
This paper talks about ImgEdit, which is a big collection of images and edits that helps AI models get better at changing pictures based on text instructions.
What's the problem?
The problem is that most AI models that edit images using text commands don't have enough good examples to learn from, and there isn't a standard way to measure how well they do.
What's the solution?
The researchers created ImgEdit, a high-quality dataset with lots of edited images and clear instructions, along with a set of rules to fairly judge how well different models perform at editing images.
Why it matters?
This matters because it helps make AI-powered image editing more accurate and reliable, which is useful for artists, designers, and anyone who wants to quickly and easily change pictures using just words.
Abstract
ImgEdit, a comprehensive image-editing dataset and benchmark, improves open-source text-to-image editing models by providing high-quality data and evaluation metrics.