imneonizer / face-app

Using single image per person to train face recognition model

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Face app

Today i am going to show you how to train your own face recognition model for free, yes we are going to use Google Colab free GPU support to train the model.

Here's the challenge

We only have single image per person

The main concept is, we will use Keras data augmentation API to generate pseudo training images, not to mention we are going to use parallel programming techniques to augment and extract embeddings faster making full use of all the free resource available on Colab.

like any other tutorials we are not going to make any dedicated colab notebook, instead we will just git clone this repository on colab and do our stuff.

Start Here Colab Notebook

Working Output

This was the final output when i was working on the project.

working


Don't forget to change your runtime type to GPU

  • Menu

runtime1

  • Hardware accelerator

runtime1

Later you can use test_model.py to recognize faces in images / videos, currently it is written for images. don't forget to have a look at this awesome library imstream for high fps video capturing.

Resources

Face Applications

About

Using single image per person to train face recognition model


Languages

Language:Jupyter Notebook 96.4%Language:Python 3.6%