Object detection, on a GPU.

The camera feed is sent to a YOLO model running on the host's GPU. Frames are processed in memory and discarded immediately — nothing is ever stored. The round-trip FPS and inference time are shown live on the HUD.