State of the art deep face analysis library

In addition to being an open source 2D&3D deep face analysis library, InsightFace also offers a range of commercial products. These include solutions for high quality face swapping and SDK development for custom applications. We are committed to providing advanced tools that drive innovation and creativity across various industries.

The library is widely used in industry, deep learning research, machine learning competitions, and open source projects.

-->
View on GitHub 22.3k

Products of InsightFace

Picsi.Ai Face Swapping

Integrated our most advanced face-swapping models: inswapper_cyn and inswapper_dax, into the InsightFace discord bot and Picsi.Ai face-swapping service, which outperform almost all similar commercial products and our open-source model inswapper_128.

See the Picsi.Ai face swapping website.

InspireFace SDK

InspireFace is a cross-platform face recognition SDK developed in C/C++, supporting multiple operating systems and various backend types for inference, such as CPU, GPU, and NPU.

See the InspireFace page.

Contact Us for Commercial Support: [email protected]

Awards

2022.08: Rank 1st of ECCV 2022 WCPA Challenge, see challenge-home.

2021.10: Rank 1st of NIST-FRVT 1:1 VISA, see visa report.

2021.04: Rank 4th of NIST-FRVT 1:1, see leaderboard.

2021.01: Rank 4th of 2020 top 10 trending AI libraries, from paperswithcode.

2019.08: Achieved 2nd place at WIDER Face Detection Challenge 2019.

2019.04: RetinaFace obtains state-of-the-art results on WIDERFace dataset.

2019.01: Achieved 1st place at iQIYI VID Challenge.

License

The code of InsightFace is released under the MIT License. There is no limitation for both academic and commercial usage. The training data containing the annotation (and the models trained with these data) are available for non-commercial research purposes only.

See the License on GitHub.

Contact Us: [email protected]

Getting started

Code of InsightFace requires Python 3.6 or higher. To install the library from PyPI run
pip install -U insightface
Open documentation

Face Recognition Projects

ArcFace

ArcFace is the state of the art face recognition approach which accepted on CVPR 2019.

See the ArcFace project page.

SubCenter-ArcFace

SubCenter-ArcFace is a face recognition approach on large-scale noisy web faces which accepted on ECCV 2020.

See the SubCenter-ArcFace project page.

VPL

VPL(Variational Prototype Learning for Deep Face Recognition) is a face recognition approach which accepted on CVPR 2021.

See the VPL project page.

Partial-FC

Partial-FC is a large-scale training framework for face recognition.

See the Partial-FC project page.

Face Detection Projects

RetinaFace

RetinaFace is the state of the art multi-tasks face detection approach which accepted on CVPR 2020.

See the RetinaFace project page.

SCRFD

SCRFD is an efficient high accuracy face detection approach.

See the SCRFD project page.

Face Alignment Projects

SDUNet

Stacked dense u-nets is a face alignment approach which accepted on BMVC 2018.

See the SDUNet project page.

CoordinateReg

CoordinateReg is an experimental face alignment approach for fast and accurate inference.

See the CoordinateReg project page.

Challenges of InsightFace

MFR Ongoing

MFR Ongoing version of ICCV-2021 Masked Face Recognition Challenge

See the InsightFace MFR Ongoing challenge page.

4th Face Anti-spoofing Workshop and Challenge, Wild Track

The 4th Face Anti-spoofing Workshop and Challenge(Wild Track) will be held in conjunction with the Computer Vision and Pattern Recognition(CVPR) 2023.

See the FAS23 challenge page.

Masked Face Recognition Challenge & Workshop ICCV 2021

The Masked Face Recognition Challenge & Workshop will be held in conjunction with the International Conference on Computer Vision (ICCV) 2021.

See the MFR challenge page.

Lightweight Face Recognition Challenge

The Lightweight Face Recognition Challenge & Workshop will be held in conjunction with the International Conference on Computer Vision (ICCV) 2019, Seoul Korea.

See the LFR challenge page.