Face Recognition App

A versatile web application build using cutting edge Face detection and recognition AI, providing you with following features :

• Face verification
• Face matching in the database
• Attendence Marking

Overview

This was the project mainly aimed to construct a robust face recognition application with wide applications using state of the art facial detection and recognition techniques. We have implemented various modules that construct the whole system, these modules are the following :

1. Face analysis module
2. Web-Application module
3. Database Module

Face Analysis Module

We designed a complete pipeline with detector backend as Retina-Face for detecting multiple faces present in the frame. We used Arc-Face and VGG-Face as models for generating face vectors. For mask detection we used Xception architecture and emotion and age detection model uses VGG-Face and Alex-net based architectures respectively.

Web-Application Module

We built a Flask Server Backend which runs the whole application by integrating all the AI models, Database, Sheet APIs and frontend HTML files and scripts together.​ We used MDBootstrap framework whic handles all the UI designs of the application.​ Javascripts for establishing the duplex communication between client and the server by accessing client-side webcam and sending snaps as post request to the backend.

Database Module

We stored facial embeddings of employees as JSON data.​ We Stored the attendance records of employees in Google Sheets.​ Used Sheets API for integration with the server.​ Finally extended the Google Sheet integration with Tableau dashboard.

An example dashboard.

Setup

Run following commands in terminal opened in the project directory to build the installation dependencies.

source init_script.sh

The above bash script -

1. Installs the requred conda environments and creates a conda environment fr-teamc
2. Setups the backend for Retinaface model
3. Downloads the pre-trained weights of Detection, Recognition, Masks, Age and Emotion models.

Here are the links for the model weights

The required models get stored in pipeline/weights file in our codebase. We provide the google drive links in case of external use -

Face Detectors

1. Retina-Face

Run the following command:

gdown --id 1oyNYwGvnCT1HOIOQq6yZ_uX06GBXzMCw

Alternatively, the link to the same is given below:

https://drive.google.com/file/d/1oyNYwGvnCT1HOIOQq6yZ_uX06GBXzMCw/view?usp=sharing

Face Recognition

1. VGG-Face

Run the following command:

gdown --id 1nuLihFS61FCGotF2KRcCzOqLhrr6wPAW

Alternatively, the link to the same is given below:

https://drive.google.com/file/d/1nuLihFS61FCGotF2KRcCzOqLhrr6wPAW/view?usp=sharing

2. ArcFace

Run the following command:

gdown --id 1atHsxw9XE1oxeipr008EkImy5n6-K-NR

Alternatively, the link to the same is given below:

https://drive.google.com/file/d/1atHsxw9XE1oxeipr008EkImy5n6-K-NR/view?usp=sharing

Age Model

Run the following command:

gdown --id 1X5c_SGcOEhfrSjvGaIYqtQk0J0sG80xu

Alternatively, the link to the same is given below:

https://drive.google.com/file/d/1X5c_SGcOEhfrSjvGaIYqtQk0J0sG80xu/view?usp=sharing

Emotion Recogntion Model

Run the following command:

gdown --id 1YPrAuQ1_CpVhloXXXa8QuTrFk5KE76Id

Alternatively, the link to the same is given below:

https://drive.google.com/file/d/1YPrAuQ1_CpVhloXXXa8QuTrFk5KE76Id/view?usp=sharing

How to use

Run the app :

conda activate fr-teamc
python app.py