Dataset Usage Rules
1.This dataset and its subsets can only be used for academic research.
2.Applicants are not allowed to use this dataset and its subsets for any commercial purposes.
3.Without permission, it is not allowed to forward, publish or distribute this dataset or its subsets to any organization or individual in any ways or by any means.
4.Any copy and sharing requests should be forwarded to the official contact email.
5.Please cite our paper if WebFace260M dataset is useful to your research:
WebFace260M: A Benchmark Unveiling the Power of Million-scale Deep Face Recognition.
Zheng Zhu, Guan Huang, Jiankang Deng, Yun Ye, Junjie Huang, Xinze Chen, Jiagang Zhu, Tian Yang, Jiwen Lu, Dalong Du, Jie Zhou.
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021.

Dataset Access
1.All users can obtain and use this dataset and its subsets only after signing the Agreement and sending it to the official contact email address.
2.For students, research groups or research institutions who need to obtain this data for academic research, signatures of both the Responsible Party and the Point of Contact are required. The Responsible Party should be the tutor of the research group or the representative of the research institution, and has the authority to be responsible for the matters related to the Agreement. (The Responsible Party should be a fixed personnel of the institution, students and other floating personnel cannot be the Responsible Party). The Point of Contact (POC) is an individual with detailed knowledge of the dataset application. In some cases, the Responsible Party and the POC may be the same person. At the same time, the homepage of the Responsible Party (or the homepage of the research group, laboratory, Department, etc., including the basic information of the Responsible Party) must be provided to prove the role of the Responsible Party.
3.Please send the signed Agreement to the official contact email ( in the form of scanned copy. Please CC the Responsible Party when sending the email.

For the participants of the WebFace260M Track of ICCV21-MFR (The Masked Face Recognition Challenge & Workshop), clean training data (WebFace42M images) and WebFace260M urls can be obtained in advance by submitting a solution exceeding the baseline results provided by us (ResNet-50, arcface loss, embedding size 512 trained by WebFace12M). The applicants which have exceeded the baseline should use the email address on codalab and send us a email which includes the agreement and the team name in the WebFace260M Track of ICCV21-MFR. For the personal participants which have exceeded the baseline but not in a research group or research institution, information in the agreement (including TITLE, EMAIL ADDRESS, The homepage of the Responsible Party) is optional. For applicants which have not exceeded the baseline, information of the agreement must be complete when applying for WebFace12M (30% WebFace42M).

Additional Information
1.At present, the open access is a subset of the complete WebFace260M dataset, which contains clean data of 600K ID and 12M images (WebFace12M). Due to the storage limitation, we provide WebFace12M in .jpg format. It is worth noting that we use .png format to train models in the original paper, so there may be some slight differences in accuracy.
2.Full and clean training data (WebFace42M) as well as WebFace260M will be released in recent time. Besides, online evaluation system will also be open. After signing the Agreement, you can obtain all the subsequent open dataset, as well as the use of online testset and evaluation system, without repeated application.
3.Official contact email address:
4.Email format for WebFace260M dataset application:
Subject: WebFace260M Dataset Application
CC: The email address of the Responsible Party
Attachment: license_agreement_for_webface260m_dataset.pdf (scanned copy)
If you want to download the WebFace260M, please send email to Copyright © XForward AI Technology Co.,LTD. 2018-2021