Running NVIDIA's LocateAnything-3B (open-vocabulary detection) fully locally from .NET, on a live camera stream

I'm the author of LocateAnythingSharp, free and open-source .NET bindings for NVIDIA's LocateAnything-3B, under the Apache 2.0 license. You type what you want to find in plain language, like "cat", "remote control" or "the red button in the corner", and you get labeled bounding boxes back. There's no fixed class list. It runs fully local. A single native engine runs the model, so there's no Python runtime and no network calls, and images never leave the machine. One build carries GPU kernels for RTX 20 through 50 series cards, and it falls back to the CPU automatically when there's no usable N