What is ID-System?

ID-System is a simple tk GUI with 2 functionalities :

1. Checking similarity between ID photo and DB embeddings

2. Checking Real Time captured picture similarity with ID photo (if ID photo embeddings 'True' in the DB)

Packages :

CUDA 11.3 Python 3.9.7 scipy 1.11.1 tqdm 4.65.0 conda 23.5.0 docker 6.1.3

There are much more packages, please check requirements.txt.

Used Models and Tools :

1. Python tk is used to make user GUI

2. RetinaFace - model to Face Detection task

3. ArcFace - model to Face Recognition

4. Google Real Time DB - database used to save photo embeddings.

5. Single-GPU RTX 3060 - used to train models

Note : All models and Tools are used with PyTorch (Python) implementation.

How .py files work ?

• There are 2 main user GUI python files: real_time_checker_gui.py and id_checker_gui.py

• Core_detector.py is the main python file with Retinaface implementation to face detection

• Core_recognizer.py is the another python file with Arcface implementation to face recognition task of Computer Vision.

• subprocess_files.py and subprocess_real_time_files.py files are responsible to connect multiple python files.

• server_connect.py file helps to connect to DB (Google Real time).

• ID_card_delete_resources.py and delete_files_real_time.py files are responsible to delete all unnecessary img and npy files. (recommended to run every time)

Note: Some codes in python files may not work or (private information) may be missed (deleted on purpose bcz of privacy. )

Folders :

• conv_npy , data, ID_picure, Real_time_image and GUI_src are folders with bunch of images (id, user real time captured image, etc.) and .npy (db saved numpy embeddings or some ArcFace calculation .npy files)

• Detector folder consists of resources of RetinaFace FD codes.

• Recognizer is another folder with Arcface FR task modules and codes.

Note: Some folders may not work or (private information) may be missed (deleted on purpose bcz of privacy. )

Workflow map :

Acknowledgements :

• RetinaFace Pytorch Implementation

• ArcFace Pytorch Implementation

• ChatGPT to search and fix bugs

• Face Anti Spoofing Silent Face

• Google Real Time Database

Citation :

Note : Most model implementation codes are open-source but have copyrights to certain authors, so I decided not to delete authors'

names in codes and also in the final similarity percentage is ~ 69 %, it is because there is no any image augimentations or pre-processing

before comaparing the image (if it is possible, try with them and hopefully you will get higher score, why not?). In the future projects of

Face Recognition, you can use that codes with modifications. I trained RetinaFace and ArcFace with my `custom dataset + WiderFace(RetinaFace)

and CelebA(ArcFace)`. So, I can't provide models and weights directly (privacy issues). If you want to use them, please contact with me first

uacoding01@gmail.com.

Thank you...