Songtao Liu, Di Huang, Yunhong Wang
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.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Object Detection | CrowdHuman (full body) | AP | 84.71 | Adaptive NMS (Faster RCNN, ResNet50) |
| Object Detection | CrowdHuman (full body) | mMR | 49.73 | Adaptive NMS (Faster RCNN, ResNet50) |
| 3D | CrowdHuman (full body) | AP | 84.71 | Adaptive NMS (Faster RCNN, ResNet50) |
| 3D | CrowdHuman (full body) | mMR | 49.73 | Adaptive NMS (Faster RCNN, ResNet50) |
| 2D Classification | CrowdHuman (full body) | AP | 84.71 | Adaptive NMS (Faster RCNN, ResNet50) |
| 2D Classification | CrowdHuman (full body) | mMR | 49.73 | Adaptive NMS (Faster RCNN, ResNet50) |
| 2D Object Detection | CrowdHuman (full body) | AP | 84.71 | Adaptive NMS (Faster RCNN, ResNet50) |
| 2D Object Detection | CrowdHuman (full body) | mMR | 49.73 | Adaptive NMS (Faster RCNN, ResNet50) |
| 16k | CrowdHuman (full body) | AP | 84.71 | Adaptive NMS (Faster RCNN, ResNet50) |
| 16k | CrowdHuman (full body) | mMR | 49.73 | Adaptive NMS (Faster RCNN, ResNet50) |