hegman12 / body_pix_tflite

This is tensorflow lite implementation of Googles Bodypix model

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

What is body_pix

A new flutter plugin project for demo of Google's Bodypix model in tensorflow lite.

Getting Started

1. Bodypix is tensorflowjs graph model. How did you convert the tensorflowjs graph model to tflite.

You can check this jupyter notebook for the conversion process. This describes how to convert tensorflowjs to tflite.

2. How to use this tflite model in android?

You can check this java file to know how to use this model. Note that this java code is not optimized. So, use this as a reference.

3. How this model perform? Can I use this for real time person masking with camera?

This is not quantized model or optimized java code. It takes around 300 to 600ms to detech 1 image as of now. With quantized model and further optimization of java code as well as environment(using custom built tflite interpretor etc) I believe its possible to improve the performance and use it in real time application with camera etc. I am using FlexDelegate and it is also adding significant overhead.

Parts segmentation result

result

About

This is tensorflow lite implementation of Googles Bodypix model

License:Apache License 2.0


Languages

Language:Jupyter Notebook 97.5%Language:Java 2.0%Language:Dart 0.3%Language:Swift 0.0%Language:Ruby 0.0%Language:Objective-C 0.0%