mudler/locate-anything.cpp ? reverse-engineered prompt

Build me a C++ app that can take an image and a text prompt, then find every matching object in the picture and return labeled boxes. I want a simple command line tool that can read a model file, run detection on CPU first, and also use GPU if it is available, then save the results as JSON and optionally write an annotated image with boxes drawn on it.

Please make it easy to build with CMake, include a basic library interface too, and add a few example commands so I can try it right away. It should support a prompt where I can name multiple things to find, like person and car, and it should work without any Python at inference time. If there is already a way to convert or quantize the model for smaller and faster runs, include that as part of the project. Use current docs online if you need to check anything.