Non-Maximum Suppression: IoU Matrix & Arc Diagram (2D)

A convolutional object detector rarely produces a single tidy box per object — it produces a cloud of overlapping candidates at slightly different positions, sizes and confidence scores, all pointing at the same thing. Instead of drawing that cloud spatially, this 2D simulator makes the algorithm's own bookkeeping visible: every candidate becomes a column ranked by confidence, and a pairwise intersection-over-union heatmap above the columns shows exactly how much any two candidates overlap. Tune the confidence threshold to prune weak proposals outright, tune the IoU threshold to control how much overlap counts as "duplicate," then run the greedy non-maximum-suppression algorithm and watch it draw an arc from each surviving column to every column it eliminates — the exact same greedy elimination running inside every real-time object detector, from YOLO to Faster R-CNN, read directly off its decision matrix instead of a rendered scene.