patlevin / face-detection-tflite

Face and iris detection for Python based on MediaPipe
MIT License
138 stars 26 forks source link

How to train your face detection model? #9

Closed vonchenplus closed 2 years ago

vonchenplus commented 2 years ago

Hello @patlevin

Thank you very much for your great work, I found a face detection model you provided (face_detection_back.tflite) is very small, I would like to ask how you trained this model? Based on what architecture or what paper?

patlevin commented 2 years ago

All models are part of Google's MediaPipe.

The face detection models are different configurations of the architecture described by the authors here. All models have been created, trained, and published by Google.

The paper describes the training process they used.

HTH, Patrick

vonchenplus commented 2 years ago

Ok thanks, I didn't find face_detection_back.tflite in mediapipe, maybe they already removed the model. Then I haven't been able to find their training code and train data.

patlevin commented 2 years ago

It seems that quite a few models used here have been removed, e.g. v0.8.7 removed the iris landmarks model.

The models used here can be found in version 0.8.1.

Then I haven't been able to find their training code and train data.

Unfortunately it is normal for training data and -code to not be available. Papers With Code is a good resource for publications that come with a public implementation.

vonchenplus commented 2 years ago

Ok thanks.