In this repository you can find the code to test and train a face detection model based on the BlazeFace architecture.

This model can run on even low-end hardware like a 2 core CPU. You can learn more about the architecture and the process to train the network in this post.

You can test the model in your computer using:

python webcam.py

The model is saved in the weights folder. You need TensorFlow 2.0, cv2, numpy and imutils installed.

To train this model you need to create a dataset:

1. Firstly, download the WIDER FACE training dataset here and the FDDB dataset here.

2. Create a folder named face_dataset and put inside the folder originalPics from the FDDB dataset and the folders from WIDER_train/images fromm the WIDER FACE training dataset.

3. Create a folder named created_dataset

To train the network we run:

python training.py --epochs=500

500 epochs are enough to converge, if you want to keep training use:

python training.py --epochs=100 --continue_training

To test the trained model:

python test_model.py

This is only going to show the first 8 images, If you want to test more images you can use the Jupyter Notebook or create a new one and copy the code from the test_model.py file.

This model is also available as an iOS app written in Swift, actually, as 2 iOS apps. The first app uses TensorFlow lite to load and run the model, and the second app uses CoreML.

You can check the post about the TensorFlow Lite version here, and the CoreML version here.