Face-heaan

1. Overview

• To implement a face ID using homomorphic encryption.

  

  Overall Structure

  • Face Identification
  • Face Registration
  • Similarity Measurement : Cosine Similarity, Manhattan Similarity, Euclidean Similarity
  • Webcam Inference

• Demo Video

2. Environment

• You must need a camera connected on your environment.
• Python version is 3.9.
• Please notice that we assume that you are using 'PyTorch' and device type as 'cpu'.
• Installing all the requirements may take some time. After installation, you can run the codes.
• Specific methods for setting preferences are described in #4.

3. Pre-Trained Model Download

• You need to download pretrained model for implementation.

• Download Link / Password : b2ec

• The pretrained model use resnet-18 without se.

• Please modify the path for the pretrained model.

• A correct location for the pre-trained model is shown below.

  ./checkpoints/resnet18_110.pth
  

4. Installation

• Please download pytorch 1.12.0 depending on your OS. Instruction is in here.

• requirements.txt file is required to set up the virtual environment for running the program. This file contains a list of all the libraries needed to run your program and their versions.

  1. In venv Environment,

  $ python -m venv [your virtual environment name]
  
  $ source [your virtual environment name]/Scripts/activate
  
  $ pip install torch==1.12.0 #for Mac
  
  $ pip install -r requirements.txt
  

  • Create your own virtual environment.
  • Write the commands to activate the virtual environment and install the necessary libraries.
  • You have a 'requirements.txt' file that lists the required libraries, use the command pip install -r requirements.txt to install libraries.

  2. In Anaconda Environment,

  $ conda create -n [your virtual environment name] python=3.9
  
  $ conda activate [your virtual environment name]
  
  $ conda install pytorch==1.12.0 -c pytorch #for Mac
  
  $ pip install -r requirements.txt
  

  • Create your own virtual environment.
  • Activate your Anaconda virtual environment where you want to install the package. If your virtual environment is named 'piheaanenv', you can type conda activate piheaanenv.
  • Use the command pip install -r requirements.txt to install libraries.

• If you encounter any conflicts, please check your dependencies carefully and reinstall according to your running environment.

5. Usage

You need to run inference_heaan.py file.

1. Face registration

   

• You can press space bar to proceed face registration.

2. Result

   
   • Unlock : When registered face is detected.
   • Lock : When registered face is not detected.
   • Too many faces : When many faces are detected.

6. Change Measurement Method

# 1) cosine similarity measurement
res_ctxt = he.cosin_sim(ctxt1, ctxt2)
result = he.compare('cosine', cos_thres, res_ctxt)

# # 2) euclidean distance measurement
# res_ctxt = he.euclidean_distance(ctxt1, ctxt2)
# result = he.compare('euclidean', euc_thres, res_ctxt)

# # 3) manhattan distance measurement
# res_ctxt = he.manhattan_distance(ctxt1, ctxt2)
# result = he.compare('manhattan', man_thres, res_ctxt)

• In inference_heaan.py file, you'll find codes like the one above. When you download and run the code, it is currently set to the cosine similarity measurement technique.
• If you want to change to the euclidean similarity measure, you can comment out the rest of the block and uncomment the second and third lines of the second block.
• If you want to change to manhattan similarity measurement, you can comment out the rest of the blocks and uncomment the second and third lines of the third block.

7. Reference

• Deng, J., Guo, J., Xue, N., & Zafeiriou, S. (2019). Arcface: Additive angular margin loss for deep face recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 4690-4699).