Home▸AI & Machine Learning▸Non-Maximum Suppression: Cleaning Up Overlapping Detection Boxes

Non-Maximum Suppression: Cleaning Up Overlapping Detection Boxes

Interactive 3D visualization of Non-Maximum Suppression: watch a greedy NMS algorithm resolve a cloud of overlapping object-detector proposal boxes into clean final detections, tuning confidence and IoU thresholds live.

AI & Machine Learning3DModerate60 FPS📱 Mobile-adapted⇄ 2D version

Object detectors don't propose one box per object — they propose dozens, jittered around every real target plus a scatter of false alarms. This simulation renders that raw proposal cloud in 3D above a ground-truth scene and runs the exact greedy Non-Maximum Suppression algorithm production detectors use to clean it up: filter by confidence, sort by score, keep the best box, discard everything that overlaps it past an IoU threshold, and repeat. Drag the confidence and IoU sliders to watch the surviving boxes change live, or generate a fresh random detector output entirely.

⚙ Under the hood

Watch the greedy Non-Maximum Suppression algorithm turn a noisy cloud of overlapping object-detector proposal boxes into clean final detections, with live confidence and IoU threshold sliders.

computer visionobject detectionNMSIoUCNNalgorithms

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install