A demonstration of Face Recognition Application on Seeed Studio reTerminal with QT5 and TensorFlow Lite. Featuring 99.3 % (LFW Validation 10-fold) accuracy facial features model and sleek Android Material design, it can be a starting point for developing Face Recognition enabled solution for Your company! It will work on Linux PC and other SBCs as well, provided necessary requirements are met.
InstallationreTerminal Raspberry Pi OS image comes with QT5 and PySide components pre-installed.
Install additional dependencies:
pip3 install -r requirements.txt
Install TensorFlow Lite Interpreter package - for faster inference install XNNPACK supported version.
Download the models and place them inside of face_rec_models directory.
Note Feature extraction model originally is from this GitHub repository, is has an accuracy of 99.3% on LFW dataset. The author didn't leave a LICENSE file in the repository, so the legal status of using pre-trained models provided for commercial purposes is undefined. You can use provided models for evaluation and development, and we're working now on fully open source models to replace them in near future.
UsageRun
DISPLAY=:0 python3 main.py
from the project main folder. Use the menu to record and delete new faces to the database.
GitHub repository
The GitHub repository for the project can be found here on Dmitry Maslov's GitHub.
Another useful link
Dmitry also has another informative tutorial on "reTerminal Machine Learning Demos (Edge Impulse and Arm NN)"
Comments