Portable YOLOv7 ONNX runtime for CPU. This is an extremely low-dependency framework that can run YOLOv7 models on a CPU. The only dependencies needed are xtensor, which is a header-only library for multi-dimensional arrays, protobuf for parsing ONNX models, and a BLAS library of your choice for matrix multiplication. Additional optional dependencies are OpenCV for drawing the bounding boxes and applying NMS (Non-Maximum Supression) on the resulting images and GoogleTest for unit testing some kernels.
I wanted to understand how inference in convolutional neural networks works and how to implement them from scratch with minimal dependencies. This was also the basis of one of my discarded Master's thesis which focused on developing a framework to run computer vision models on RISC-V CPUs.
Most likely not!