Code for the paper "An End-to-End Optimized Lensless System for Privacy-Preserving Face Verification".

git clone https://github.com/OpenImagingLab/LenslessFace.git cd LenslessFace conda create -n lenslessface python=3.9 conda activate lenslessface pip install -r requirements.txt 

For training, we use the Asian-Celeb dataset. Tests are conducted on the LFW dataset and FCFD dataset , which should be downloaded and extracted to the data directory.

You can modify the arguments in config_file to change the dataset path.

For RGB-based teacher model training, run the following command:

./scripts/dist_train_teacher.sh config_file 

An example of config_file is configs/face_no_optical/rgb_teacher.py.

For lensless-based student model training, run the following command:

./scripts/dist_train.sh config_file 

An example of config_file is configs/distill/face/base.py.

For lensless-based face center detection model training, run the following command:

./scripts/dist_train_pose.sh config_file 

An example of config_file is configs/face_center_detection/base.py.

For aligned face verification, run the following command:

./scripts/test.sh config_file check_point_path 

An example of config_file is configs/distil/face/base.py.

For random face verification, run the following command:

./scripts/test_random.sh config_file 

An example of config_file is configs/hybrid/optical/base_test.py. you need to modify the cls_checkpoint and face_center_detection_checkpoint in the config_file.

We thank the authors and maintainers of the following repositories for providing the frameworks and datasets that significantly facilitated our research:

• OpenMMLab

• LenslessClassification

• LenslessPiCam

Special thanks also go to the authors of the datasets we used for training and evaluation.

Citation Please cite our paper if you find this repository useful for your research:

@misc{cai2024lenslessface, title={LenslessFace: An End-to-End Optimized Lensless System for Privacy-Preserving Face Verification}, author={Xin Cai and Hailong Zhang and Chenchen Wang and Wentao Liu and Jinwei Gu and Tianfan Xue}, year={2024}, eprint={2406.04129}, archivePrefix={arXiv}, primaryClass={cs.CV} } 

This project is licensed under the terms of the MIT license.