Image segmentation on an STM32H7 STM32H7 microcontroller by implementing a UNet-inspired network. The network is designed and trained in Tensorflow and then deployed in C using STM32's X-Cube-AI library to import a .h5 file
The largest output of any layer is only 262 Kb and the board has 564 Kb of RAM. It runs inference at around 3-4 FPS by transfering images from PC to the MCU over USB.
Project Video Demo
Project Report
Dataset Link, (search Portseg_128)
Dataset Link if the previous link does not work
