Hello world activity for STM32CubeIDE
This project includes information for training a model in tensorflow, converting it to TF Lite, and running inference on an STM32L476RGx board.
Takeaways:
Pros
- Compared to TF Lite for Microcontrollers, xCube AI uses about 40% less flash space, and is about 25% faster (for this project)
- Out of the different libraries explored, cubeIDE has been the easiest to work with
- Can easily accept pre-trained models from keras and tflite (tf version cannot be newer than 2.2 for some reason)
Cons
- xCube AI is closed source
- xCube used about 4% more RAM than TF Lite for microcontrollers (for this project)
NOTICE:
I did not design or create this exercise, most of this project was learned from various tutorials created by Digi-Key and Tensorflow.