.png){kind=icon}
.png){kind=icon}
Return
ANMS: attention-based non-maximum suppression
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-022-12142-5.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Non-Maximum Suppression (NMS) is an essential part of the object detection pipeline. However, due to the inconsistency between the classification confidence and the object localization, NMS may mistakenly eliminate the bounding boxes with low classification confidence and high localization accuracy. In this paper, we propose an attention-based non-maximum suppression (ANMS) algorithm. It reconstructs the attention map to obtain the object location information by backpropagating the top-level object classification semantic information. Furthermore, integrating the classification confidence and the attention map of the detection bounding boxes adjust the inconsistency between the classification confidence and the object localization. On the PASCAL VOC2007 and the PASCAL VOC2012 datasets, the proposed ANMS algorithm achieved 1.85 and 1.24 performance improvement over the NMS algorithm. On the MS COCO datasets, the proposed ANMS algorithm achieved 0.3 performance improvement, which proved the ANMS algorithm's effectiveness.
Keywords:
Object detection
Non-maximum suppression
Attention map
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W