NMS Maker

NMS Maker provides a structured approach to applying non-maximum suppression on detector results. It accepts a list of candidate bounding boxes with confidence scores and class labels, and outputs a pruned set of detections. Conceptually, it sorts detections by score, computes pairwise overlaps via IoU, and removes or downscales less-preferred boxes when overlaps exceed the configured threshold. The tool supports standard NMS and optional soft-NMS, per-class suppression, and batched inputs for large datasets. It does not train models or alter underlying predictions, but it optimizes post-processing to improve precision and reduce duplicate detections. Use cases include evaluating object detectors on COCO- and VOC-style datasets, benchmarking post-processing strategies, and deploying consistent suppression rules across experiments.