BriefGPT.xyz
Jul, 2018
获取定位置信度以实现准确目标检测
Acquisition of Localization Confidence for Accurate Object Detection
HTML
PDF
Borui Jiang, Ruixuan Luo, Jiayuan Mao, Tete Xiao, Yuning Jiang
TL;DR
本文提出了一种CNN方法，名为IoU-Net，以预测物体检测边界框的IoU值，使得模型具有定位置信度，进而提高非极大值抑制算法（NMS）的精度，在MS-COCO数据集上实验表明IoU-Net方法适用于多种先进的物体检测器且有效。
Abstract
Modern
cnn
-based object detectors rely on
bounding box regression
and non-maximum suppression to localize objects. While the probabilities for class labels naturally reflect classification confidence,
→