A fast C++ ggml inference port of NVIDIA LocateAnything-3B for open-vocabulary detection on CPU and GPU.