[COSCUP 2018] uTensor C++ Code Generator

uTensor is a neural network inference library for microcontrollers (MCUs) that automatically generates C++ code from trained TensorFlow models. It includes a command line tool that takes a trained TensorFlow model file as input and outputs C++ header and source files to run the model on MCUs. The tool currently supports generating code for multilayer perceptrons (MLPs) and convolutional neural networks (CNNs). The presentation demonstrated an end-to-end example of training a CNN model in TensorFlow, saving it as a protocol buffer file, and using the uTensor tool to generate C++ code to run inference on the model.