A new flutter plugin project for demo of Google's Bodypix model in tensorflow lite.
You can check this jupyter notebook for the conversion process. This describes how to convert tensorflowjs to tflite.
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.
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.