This is the offical repository of the paper:

Unsupervised Face Recognition using Unlabeled Synthetic Data

Arxiv

Paper accepted at Face and Gesture 2023

Pretrained Models

Model	Images	LFW	AgeDB-30	CFP-FP	CA-LFW	CP-LFW	Pretrained Model
USynthFace	100K	92.12	71.08	78.19	76.15	71.95	download
USynthFace	200K	91.93	71.23	78.03	76.73	72.27	download
USynthFace	400K	92.23	71.62	78.56	77.05	72.03	download

Requirements

Requirements for DiscoFaceGAN Image Generation:

• Python 3.6
• Tensorflow 1.12 with GPU support

We recommend creating a virtual environment with requirementsTF.txt.
Download pretrained DiscoFaceGAN, strickly follow DiscoFaceGAN license and save in DiscoFaceGAN/pretrained/.

Requirements for USynthFace Training

• pytorch 1.11.0
• torchvision 0.12.0

We recomment creating a virtual environment with requirementsTorch.txt

Training Dataset Preparation

To generate images run in DiscoFaceGAN/:

generate_imgs.sh --save_path "save/path/of/unaligned/images"

To align images run:

align_imgs.sh --in_folder "path/to/image/folder" --out_folder "save/path/of/aligned/images"

Set datapath="../.." in config/config.py to folder with aligned DiscoFaceGAN images.

Evaluation Dataset Preparation

Download evaluation datasets from insightface in strict compliance with the license distribution. Evaluation datasets are available e.g. in the training dataset package CASIA-Webface as bin files.
Set eval_datasets="../.." in config/config.py to your unzipped folder which includes the bin files.

Train USynthFace

Change config/config.py and train.sh to your preferences and execute:

train.sh

To reproduce the results of the pretrained models, change number_of_images= and output_dir= in config/config.py.

Evaluate USynthFace

In evaluation/ run:

CUDA_VISIBLE_DEVICES=0 python eval.py --model_folder "path/to/model/folder/" --rec_path "path/to/folder/with/bin/files"

Test log is saved in model_folder.

References:

• DiscoFaceGAN
• Moco

If you use any of the code provided in this repository, please cite the following paper:

Citation

@inproceedings{DBLP:conf/fgr/BoutrosKFKD23,
  author    = {Fadi Boutros and
               Marcel Klemt and
               Meiling Fang and
               Arjan Kuijper and
               Naser Damer},
  title     = {Unsupervised Face Recognition using Unlabeled Synthetic Data},
  booktitle = {17th {IEEE} International Conference on Automatic Face and Gesture
               Recognition, {FG} 2023, Waikoloa Beach, HI, USA, January 5-8, 2023},
  pages     = {1--8},
  publisher = {{IEEE}},
  year      = {2023},
  url       = {https://doi.org/10.1109/FG57933.2023.10042627},
  doi       = {10.1109/FG57933.2023.10042627},
}

License

This project is licensed under the terms of the Attribution-NonCommercial-ShareAlike 4.0 
International (CC BY-NC-SA 4.0) license. 
Copyright (c) 2021 Fraunhofer Institute for Computer Graphics Research IGD Darmstadt