Adaptive NMS: Refining Pedestrian Detection in a Crowd

TL;DR

This paper introduces adaptive-NMS, a dynamic suppression method for pedestrian detection in crowds, which improves bounding box refinement by considering crowd density, leading to state-of-the-art results.

Contribution

We propose a novel adaptive-NMS algorithm with a density-aware threshold and a learnable density scoring subnetwork, enhancing pedestrian detection in crowded scenes.

Findings

Achieved state-of-the-art results on CityPersons and CrowdHuman benchmarks.

Demonstrated improved detection accuracy in crowded pedestrian scenarios.

Efficient integration of density scoring into existing detectors.

Abstract

Pedestrian detection in a crowd is a very challenging issue. This paper addresses this problem by a novel Non-Maximum Suppression (NMS) algorithm to better refine the bounding boxes given by detectors. The contributions are threefold: (1) we propose adaptive-NMS, which applies a dynamic suppression threshold to an instance, according to the target density; (2) we design an efficient subnetwork to learn density scores, which can be conveniently embedded into both the single-stage and two-stage detectors; and (3) we achieve state of the art results on the CityPersons and CrowdHuman benchmarks.