hyprnx / face-recognition

Face Recognition on Jetson Nano B01

Face Recognition on Jetson Nano

Table of contents

• About The Project
• Getting Started
  • Prerequisites
  • Installation
• Details
  • Model
  • Results
  • Performance
• Challenges
• License

About The Project

I deployment face recognition on jetson nano and build a database to load and record the attendance of people.

Getting Started

To get a local copy up and running follow these simple steps.

Prerequisites

• Use LXDE as your Desktop Environment
• Extent swap partitions
• Install torch2trt

Installation

1. Clone the repo

   git clone https://github.com/phamvanhanh6720/Face-Recognition.git

2. Install requirements

   pip3 install -r requirments.txt

Details

Model

• I use the pre-trained model to extract embedding of a face image and use cosine similarity between extracted embedding and other stored embeddings in the system. After that, I choose a threshold to identify the person.

Results

• My model has the ability to distinguish real or fake faces based on 2D images. However, my model is sensitive when the user uses a color image to fool the system.
• I have not evaluated my model on a large test set, but which achieves good accuracy on the small test set.

Performance

• A full pipeline of process a frame contains face takes approximately 0.2s. So my system achieves 5 FPS

Challenges

• There are several challenges when you deploy on jetson nano:
  • Memory. If you run model with cuda, it takes so much memory
  • Installing few libraries is slow and hard

License

Distributed under the MIT License. See LICENSE for more information.