https://arxiv.org/abs/1711.11575 

Abstract

1.Introduction

2.Related Works

　　Object relation in post-processing

　　Sequential relation modeling

　　Human centered scenarios

　　Duplicate removal

　　Attention modules in NLP and physical system modeling

3.Object Relation Module

　　The relation feature 

　　The relation weight 　　

　　Apperaance weight　　

　　Geometry weight 　　

 4.Relation Networks For Object Detection

4.1.Review of Object Detection Pipeline

　　First step generates full image features.

　　Second step generates regional features.

　　Third step performs instance recognition.

　　Last step performs duplicate removal.

4.2.Relation for Instance Recognition

4.3.Relation for Duplicate Removal

Notes for "Relation Networks for Object Detection"

原文地址：https://www.cnblogs.com/vincent1997/p/9870587.html