HQ-Edit is a high-quality, instruction-based image editing dataset created by a research team at the University of California, Santa Cruz. This dataset contains approximately 200,000 editing examples, each with an input image, an output image, and detailed editing instructions. The dataset is characterized by high resolution, diverse editing instructions, and accurately aligned images and text. Using the latest base models GPT-4V and DALL-E 3, the researchers built a scalable data collection process that automatically generates high-quality image editing data. The high-resolution images and rich details in the HQ-Edit dataset significantly improve the performance of existing image editing models.

Citation

@article{hui2024hq,
title   = {HQ-Edit: A High-Quality Dataset for Instruction-based Image Editing},
author  = {Hui, Mude and Yang, Siwei and Zhao, Bingchen and Shi, Yichun and Wang, Heng and Wang, Peng and Zhou, Yuyin and Xie, Cihang},
journal = {arXiv preprint arXiv:2404.09990},
year    = {2024}
}