sod

Why we included this project

SOD is a C library aimed at devices where a GPU isn't available. It bundles classic image processing with convolutional and recurrent networks for real-time multi-class object detection, and includes its own lightweight RealNet architecture for tasks like face detection and facial landmark extraction. Everything ships as a single amalgamated C file with no dependency tree, so dropping it into a small project or embedded target is straightforward, and it plays nicely with OpenCV if you already use that. The CPU-focused design also means you can train RealNet models on a modern desktop processor, which is handy for IoT cameras or other embedded appliances that need pre-trained models and a compact API.