Self Contained TFLite Hello World Project for Raptor (efabless)

This repo explains the intimitading TensorFlow repo and demonstrates TFLite on raptor by making it blinking.

How to Build

After cloning this repo and cd to it, run the following command:

make -f tensorflow/lite/micro/tools/make/Makefile TARGET=raptor hello_world_bin

Parameters:

• -f tensorflow/lite/micro/tools/make/Makefile: the path to the Makefile.
• TARGET: the target MCU, in this case raptor
• hello_world_bin: the project name (hello_world) + the build type, and in this case we want the binary (bin)

Porting Any Platform to TFLite

The Tensorflow repo might seem intimitading when you are just starting to experiment with it. Therefore, I created this repo which is a simpler version of if while maintaining the original structure.

When writing TFLite project for MCUs, it is essential that you understand the following directories in TF:

1. Start with the Hello World example and modify it as you want to make it fit your application purpose. It is found in the following path tensorflow/lite/micro/examples/hello_world
2. If you have (and most probably you will) platform specific implementations (such as ADC data collection, DAC, UART, etc.) needed in the application, place these implementations in a subdirectory with the platform name inside the hello_world folder.
   example: hello_world/raptor/ contains the raptor version of output_handler.cc
   
3. Provide your platform's implementation of the tensorflow/lite/micro/<platform_name>/debug_log.cc, here <platform_name> is raptor. This file provides a way to output debug information (for example using the UART).
4. Write a Makefile for your platform and put it in: tensorflow/lite/micro/tools/make/targets/<platform_name>_makefile.inc For example, you will find raptor's Makefile in tensorflow/lite/micro/tools/make/targets/raptor_makefile.inc
5. Finally, build using the command:

make -f tensorflow/lite/micro/tools/make/Makefile TARGET=<platform_name> hello_world_bin