A Convnet for Non-maximum Suppression

By Jan Hendrik Hosang, Rodrigo Benenson, Bernt Schiele. 2015

78 citations recorded by the source corpus.

Non-maximum suppression (NMS) is used in virtually all state-of-the-art object detection pipelines. While essential object detection ingredients such as features, classifiers, and proposal methods have been extensively researched surprisingly little work has aimed to systematically address NMS. The de-facto standard for NMS is based on greedy clustering with a fixed distance threshold, which forces to trade-off recall versus precision. We propose a convnet designed to perform NMS of a given set of detections. We report experiments on a synthetic setup, crowded pedestrian scenes, and for general person detection. Our approach overcomes the intrinsic limitations of greedy NMS, obtaining better recall and precision.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Learning Non-maximum Suppression (2017, methodology): “Since we want to avoid the discretisation scheme used in [13], we will solve this issue with pooling across detections (the second key).”
Spatial Memory for Context Reasoning in Object Detection (2017, methodology): “To bypass NMS, LSTM [35] cells arranged in 2D order [78] and intersection-over-union (IoU) maps [39] have been used for single-class object detection.”
Detection in Crowded Scenes: One Proposal, Multiple Predictions (2020, background): “Other works such as [23, 17] propose to predict different NMS thresholds for different bounding boxes.”
Improving Object Localization with Fitness NMS and Bounded IoU Loss (2017, unknown): “he greatest conﬁdence in the current class. Recent work has proposed decaying the associated detection score (as opposed to discarding detections) [1] and learning the NMS function directly from data [7] [8], however in the following we investigate the score(:) and same(:) functions without changing”