Critic-10K Dataset

This repository hosts the Critic-10K dataset, introduced in the paper The Consistency Critic: Correcting Inconsistencies in Generated Images via Reference-Guided Attentive Alignment.

The Critic-10K dataset is specifically constructed to address and rectify inconsistencies in generated images. It comprises reference-degraded-target triplets, obtained through VLM-based selection and explicit degradation. This dataset effectively simulates common inaccuracies or inconsistencies observed in existing image generation models, making it ideal for research into reference-guided post-editing approaches.

• Project Page: https://ouyangziheng.github.io/ImageCritic-Page/
• Code: https://github.com/HVision-NKU/ImageCritic

Dataset Download

You can download the Critic-10K training dataset directly from Hugging Face.

Download via Python (from the ImageCritic GitHub repository)

python /raid/users/oyzh/ImageCritic/download_dataset.py

Note: This Python script is typically found in the ImageCritic GitHub repository and is designed to download the dataset.

Download via Git

git lfs install
git clone https://huggingface.co/datasets/ziheng1234/Critic-10K

Citation

If you find this project useful for your research, please consider citing our paper:

@article{ouyang2025consistency,
  title={The Consistency Critic: Correcting Inconsistencies in Generated Images via Reference-Guided Attentive Alignment},
  author={Ouyang, Ziheng and Song, Yiren and Liu, Yaoli and Zhu, Shihao and Hou, Qibin and Cheng, Ming-Ming and Shou, Mike Zheng},
  journal={arXiv preprint arXiv:2511.20614},
  year={2025}
}

License

This dataset is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0) for non-commercial use only. Any commercial use requires formal permission.