.png){kind=icon}
.png){kind=icon}
Return
Dynamic sample weighting for weakly supervised object detection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.imavis.2022.104444.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The framework based on Multiple Instance Learning (MIL) greatly improves the performance of Weakly Supervised Object Detection (WSOD), which enjoys a promising development. However, the detection results tend to be the most discriminative parts of the object, which is still an open problem. In this paper, we analyze the causes of the problem from the perspective of sample balance. Considering the inaccuracy of pseudo supervised information in WSOD, a Dynamic Sample Weighting strategy (DSW) is proposed to focus on samples which closely cover the object, making the detection results cover the object more comprehensively. The performance of DSW on PASCAL VOC 2007, PASCAL VOC 2012 and MS-COCO is significantly enhanced through the simple and effective method in this paper. Code will be made available.(c) 2022 Elsevier B.V. All rights reserved.
Keywords:
Weakly supervised learning
Object detection
Dynamic sample weighting
Multiple instance learning