Object-detection results drawn over the picture they came from — per-class tinted boxes with label and confidence chips, a live confidence gate, a class legend, and a rail whose hover, focus and selection stay in sync with the boxes.