Dataset Card for HQ-EDIT

HQ-Edit, a high-quality instruction-based image editing dataset with total 197,350 edits. Unlike prior approaches relying on attribute guidance or human feedback on building datasets, we devise a scalable data collection pipeline leveraging advanced foundation models, namely GPT-4V and DALL-E 3. HQ-Edit’s high-resolution images, rich in detail and accompanied by comprehensive editing prompts, substantially enhance the capabilities of existing image editing models.

• Homepage: https://thefllood.github.io/HQEdit_web/
• Repository: https://github.com/UCSC-VLAA/HQ-Edit

If you would like to preview the data online using Dataset Viewer, please visit:

• Dataset Demo: https://huggingface.co/datasets/UCSC-VLAA/HQ-Edit-data-demo

Dataset Structure

"input" (str): description of input image.

"input_image" (image): the input image.

"edit" (str): edit instruction for transforming input images to output images.

"inverse_edit" (str): inverse-edit instructions for transforming output images back to input images.

"output" (str): description of output image.

"output_image" (image): the output image.

Citation

If you find this dataset useful, please consider citing our paper:

@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}
}