Meng-Jiun Chiou, Henghui Ding, Hanshu Yan, Changhu Wang, Roger Zimmermann...
TL;DR本文提出一种基于 PU 学习和动态标签频率估计的方法,通过有效利用标签频率从而解决场景图生成中的长尾问题和报告偏差,实现更准确的标签估计和更平衡的场景图生成。
Abstract
Given input images, scene graph generation (SGG) aims to produce
comprehensive, graphical representations describing visual relationships among
salient objects. Recently, more efforts have been paid to the long tail pro