CELEB-500K

official github for paper "CELEB-500K: A LARGE TRAINING DATASET FOR FACE RECOGNITION"

Image urls

link: https://pan.baidu.com/s/1Cc0InuxCoWA480r1pCNYug password: hseu

Usage

Download image urls from the link above and run DownloadImages.py to download images.
Then, face detection and alignment should be applied according to your own interest.
We suggest label cleaning for 2-3 rounds as described in the paper to obtain a cleaned dataset.

Reference

@inproceedings{cao2018celeb,
title={Celeb-500K: A Large Training Dataset for Face Recognition},
author={Cao, Jiajiong and Li, Yingming and Zhang, Zhongfei},
booktitle={2018 25th IEEE International Conference on Image Processing (ICIP)},
pages={2406--2410},
year={2018},
organization={IEEE}
}