BriefGPT.xyz
Oct, 2018
视觉问答系统的注意力分析
Knowing Where to Look? Analysis on Attention of Visual Question Answering System
HTML
PDF
Wei Li, Zehuan Yuan, Xiangzhong Fang, Changhu Wang
TL;DR
本文结合注意力机制提出了两种最先进的视觉问答方法,并通过可视化和分析它们的估计注意力图来研究它们的鲁棒性和缺点。研究表明两种方法对特征敏感,同时对于计数和多对象相关的问题表现不佳。该研究结果和分析方法可帮助研究人员识别重要的挑战,以改进自己的VQA系统。
Abstract
attention mechanisms
have been widely used in
visual question answering
(VQA) solutions due to their capacity to model deep cross-domain interactions. Analyzing
→