Face Recognition with Arcface

Arcface is a efficient and effective face recognition algorithm, which can train on tens of millions identity on a single server. Real-time face recognition in unconstrained environments. Using Shuffle Attention MobileNetV3 Architecture as Backbone, and modified ArcFace as loss function.

This module can get a number of names as input for tracking the specific face.

Experiments

on LFW

Attempt	Parameters	Madds	Top1-acc
Mobile Face Net
SA-MobileNetV3-Large with modified ArcFace loss

Installation

1- Install pytorch (torch>=1.6.0).

2- install requirements

pip install -r requirements.txt

3- MobileNet weights exist in ./weights directory. if you want ResNet weights, download with this command:

python download_weights.py

Dataset

Download the dataset from here.

Train

To train a model, run train.py with the path to the configs. for example if you want train SA-MobileNetV3, run this command:

python train.py configs/webface_samnv3

Train on multi GPUs:

python -m torch.distributed.launch --nproc_per_node=4 --nnodes=1 --node_rank=0 train.py configs/webface_samnv3

other configs are available in ./configs

Test

Run the following command for evaluation trained model on test dataset:

python ...

Predict

Run the following command for classification images:

python predict.py --input input/sajjad_0.jpg

Inference

For feature extraction, run this command:

python inference_compare.py --input1 input/sajjad_0.jpg --input2 input/sajjad_1.jpg

Put your input images or videos in ./input directory. The output will be saved in ./output. In root directory of project, run the following command:

python inference_video.py --input "./input/sample.mp4" --update

or

python inference_image.py --input "./input/sajjad.jpg" --update

Use -sh for representation of results during code running or not

Note that you can pass some other arguments. Take a look at --help argument for each command.

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}
}