uTensor

uTensor is an extremely lightweight machine learning inference framework built on TensorFlow and optimized for Arm targets. It consists of a runtime library and an offline tool that translates trained TensorFlow models into C++11 code (.cpp and .hpp files) for embedded inferencing. The core runtime is only about 2KB in size, with a memory footprint of approximately 1275 bytes of text, 4 bytes of data, and 28 bytes of BSS. The framework emphasizes system safety through dedicated memory regions for tensor metadata and data, eliminating heap sharing and enabling safe remote model updates. It also prioritizes compile-time error checking to catch mismatched inputs, outputs, and invalid memory accesses. uTensor includes tutorials for error handling and custom operators, and its rearchitecture focuses on low power consumption, small dynamic footprint, and debuggable code.