A field of candidate bounding boxes over a stylized scene, filtered live by a confidence threshold and a real IoU (Intersection-over-Union) non-max-suppression pass down to the final detections.
Every candidate box carries a class and a confidence score; the confidence slider removes low-confidence guesses, and the IoU slider controls how aggressively overlapping boxes of the same class get merged down to one detection per object.
Pick a scene, drag the confidence and IoU sliders to watch the filter recompute instantly, adjust detection speed, and use Rebuild or Play/Pause to generate a fresh scatter of candidate boxes.
Non-max suppression is what turns YOLO's dozens of overlapping candidate boxes per object into a single clean detection — without it, every real object would be reported many times over.