Official repository for MixFaceNets: Extremely Efficient Face Recognition Networks

Last update: Dec 13, 2022

Overview

MixFaceNets

This is the official repository of the paper: MixFaceNets: Extremely Efficient Face Recognition Networks.

(Accepted in IJCB2021) https://ieeexplore.ieee.org/abstract/document/9484374

Model	MFLOPs	Params (M)	LFW%	AgeDB-30%	IJB-B( TAR at FAR1e–6)	IJB-C( TAR at FAR1e–6)	Pretrained model
MixFaceNet-M	626.1	3.95	99.68	97.05	91.55	93.42	pretrained-mode
ShuffleMixFaceNet-M	626.1	3.95	99.60	96.98	91.47	93.5	pretrained-mode
MixFaceNet-S	451.7	3.07	99.60	96.63	90.17	92.30	pretrained-mode
ShuffleMixFaceNet-S	451.7	3.07	99.58	97.05	90.94	93.08	pretrained-mode
MixFaceNet-XS	161.9	1.04	99.60	95.85	88.48	90.73	pretrained-mode
ShuffleMixFaceNet-XS	161.9	1.04	99.53	95.62	87.86	90.43	pretrained-mode

FLOPs vs. performance on LFW (accuracy), AgeDB-30 (accuracy), MegaFace (TAR at FAR1e-6), IJB-B (TAR at FAR1e-4), IJB-C (TAR at FAR1e-4) and refined version of MegaFace, noted as MegaFace (R), (TAR at FAR1e-6). Our MixFaceNet models are highlighted with triangle marker and red edge color.

LFW

AgeDb-30

MegaFace

MegaFace(R)

IJB-B

IJB-C

If you find MixFaceNets useful in your research, please cite the following paper:

Citation

@INPROCEEDINGS{9484374,
  author={Boutros, Fadi and Damer, Naser and Fang, Meiling and Kirchbuchner, Florian and Kuijper, Arjan},
  booktitle={2021 IEEE International Joint Conference on Biometrics (IJCB)}, 
  title={MixFaceNets: Extremely Efficient Face Recognition Networks}, 
  year={2021},
  volume={},
  number={},
  pages={1-8},
  doi={10.1109/IJCB52358.2021.9484374}}

The model is trained with ArcFace loss using Partial-FC algorithms. If you train the MixfaceNets with ArcFace and Partial-FC, please follow their distribution licenses.

Citation

@inproceedings{deng2019arcface,
  title={Arcface: Additive angular margin loss for deep face recognition},
  author={Deng, Jiankang and Guo, Jia and Xue, Niannan and Zafeiriou, Stefanos},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={4690--4699},
  year={2019}
}
@inproceedings{an2020partical_fc,
  title={Partial FC: Training 10 Million Identities on a Single Machine},
  author={An, Xiang and Zhu, Xuhan and Xiao, Yang and Wu, Lan and Zhang, Ming and Gao, Yuan and Qin, Bin and
  Zhang, Debing and Fu Ying},
  booktitle={Arxiv 2010.05222},
  year={2020}
}

Official repository for MixFaceNets: Extremely Efficient Face Recognition Networks

Related tags

Overview

MixFaceNets

This is the official repository of the paper: MixFaceNets: Extremely Efficient Face Recognition Networks.

Citation

Citation

Owner

Fadi Boutros

Publication describing 3 ML examples at NSLS-II and interfacing into Bluesky

Face Mask Detection is a project to determine whether someone is wearing mask or not, using deep neural network.

Vis2Mesh: Efficient Mesh Reconstruction from Unstructured Point Clouds of Large Scenes with Learned Virtual View Visibility ICCV2021

ImageNet Adversarial Image Evaluation

Learning Chinese Character style with conditional GAN

In this repo we reproduce and extend results of Learning in High Dimension Always Amounts to Extrapolation by Balestriero et al. 2021

Large-Scale Unsupervised Object Discovery

Keep CALM and Improve Visual Feature Attribution

SeqTR: A Simple yet Universal Network for Visual Grounding

Here I will explain the flow to deploy your custom deep learning models on Ultra96V2.

Kaggle | 9th place (part of) solution for the Bristol-Myers Squibb – Molecular Translation challenge

This repo provides the source code & data of our paper "GreaseLM: Graph REASoning Enhanced Language Models"

FusionNet: A deep fully residual convolutional neural network for image segmentation in connectomics

TagLab: an image segmentation tool oriented to marine data analysis

This repository lets you interact with Lean through a REPL.

Mask-invariant Face Recognition through Template-level Knowledge Distillation

Optimizing DR with hard negatives and achieving SOTA first-stage retrieval performance on TREC DL Track (SIGIR 2021 Full Paper).

A customisable game where you have to quickly click on black tiles in order of appearance while avoiding clicking on white squares.

Pixel-level Crack Detection From Images Of Levee Systems : A Comparative Study

This solves the autonomous driving issue which is supported by deep learning technology. Given a video, it splits into images and predicts the angle of turning for each frame.