logo
CanyamAI summaries for academic research
language
MobileMobile
  • Phone FeatureWeChat Mini Program
EnglishEnglish
noticeNotice
BlogBlog
delete
arrow
Return

A weakly-supervised oriented object detector : Knowledge-based dropblock and unified regression network

delete2025-07-06
delete0
PRE
AI
段立娟 cover
段立娟 (Lijuan Duan)
Z
Zichen Zhang
Z
Zhaoying Liu
F
Fengjin Xiao
DOI：10.1016/j.neunet.2025.107830delete
deleteOriginal
deleteOriginal request for help
deleteShare
deleteSave
Abstract

Abstract

En 中文
• We proposed a knowledge-based dropblock and unified regression network (KDUNet), which improves the global information extraction and precise location capability of weakly-supervised network based on horizontal annotation. • We designed a large-view background attention block to refine feature maps in knowledge-based dropblock (KBD). This facilitates selective hiding of object identity parts and highlighting the overall object area. • We developed a comprehensive unambiguous unified project (UUP) mechanism between RBox and HBox which resolves potential ambiguities introduced by different spatial transformations.

Journal

Neural Networks cover
Neural Networks
IF:
6.3
Papers:
7.8K
Citations:
3.0W

Organization

No organization information available