WebFace260M

News

December-31-2021: The WebFace260M Benchmark has been re-opened and moved to the new Codalab website here.

October-15-2021: The final leaderboard for WebFace260M Track includes ‘Main Result’ and ‘SFR Result’ is available here.

August-19-2021: The official report of WebFace260M Track in the ICCV-21 MFR challenge is available on arxiv. The challenge is under going till October 1, 2021.

June-7-2021: We organize the the Masked Face Recognition Challenge & Workshop (MFR) in ICCV 2021. For participants of WebFace260M Track, we provide the full WebFace260M data in advance. Details are in this link.

March-9-2021: A preprint of WebFace260M benchmark paper is available on arxiv.

March-8-2021: The benchmark website is online.

March-1-2021: Our WebFace260M benchmark paper is accepted by CVPR 2021.

What is WebFace260M?

WebFace260M is a new million-scale face benchmark, which is constructed for the research community towards closing the data gap behind the industry.

Noisy 4M identities and 260M faces

High-quality training data with 42M images of 2M identities by using automatic cleaning

A test set with rich attributes and a time-constrained evaluation protocol

Sponsors

Collaborators

Zheng Zhu* THU, XForwardAI
Guan Huang* XForwardAI
Jiankang Deng Imperial College London
Yun Ye XForwardAI
Junjie Huang XForwardAI
Xinze Chen XForwardAI
Jiagang Zhu XForwardAI
Tian Yang XForwardAI
Jiwen Lu THU
Dalong Du XForwardAI
Jie Zhou THU

Highlights

Noisy 4M identities and 260M images (WebFace260M)

High-quality 2M identities and 42M images (WebFace42M)

ArcFace models (ResNet-100) trained on the WebFace42M achieve state of the art on IJBC

The statistics of our test set

Only based on the WebFace42M, XForwardAI achieves overall 3rd on 1:1 verification among 430 submissions, on the leaderboard of NIST-FRVT in October 2020

Using WebFace42M as foundation, XForwardAI further ranks 1st on Face Mask Effects with absolute advantage, on the latest leaderboard of NIST-FRVT

Using the same model, XForwardAI also achieves overall 3rd on 1:1 verification among 492 submissions, on the latest leaderboard of NIST-FRVT

Paper

Please cite the following paper if WebFace260M is useful to your research:

              @inproceedings {zhu2021webface260m,
                title=  {WebFace260M: A Benchmark Unveiling the Power of Million-scale Deep Face Recognition},
                author=  {Zheng Zhu, Guan Huang, Jiankang Deng, Yun Ye, Junjie Huang, Xinze Chen,
                    Jiagang Zhu, Tian Yang, Jiwen Lu, Dalong Du, Jie Zhou},
                booktitle=  {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
                year=  {2021}              
                 }