论文阅读：Contextual Translation Embedding for Visual Relationship Detection and SGG(PAMI2020)

还是论文题目太长打不下了（SGG：场景图生成）

中心思想：p≈u-s-o

1.目标检测

2.视觉特征提取
出于对论文上下文的理解，我觉得这里的主客体特征应该融合了fasterrcnn提取的视觉特征和主客体的位置特征

3.谓语特征=union box特征-主语特征-宾语特征

4.谓语特征经过两层fc得到视觉模块的谓语类别的置信度
同时4’.谓语特征经过一层fc，主语和宾语分别做word embedding。

5.在三个连续的时间步内分别以主语特征、谓语特征、宾语特征为输入，得到它们的hidden state，再经过一层fc得到语言模块的谓语类别分数

6.最终三元组的分数：

zs,zo：主客体的置信度
zp：视觉模块的置信度
zl：语言模块的置信度

---------------------------一些碎碎念---------------------------
没啥说的了。
大道至简就完事了。

论文阅读：Contextual Translation Embedding for Visual Relationship Detection and SGG(PAMI2020)相关推荐

CVPR2020 | 论文阅读——Multiple Anchor Learning for Visual Object Detection
MAL 用于视觉目标检测的多锚点学习 Abstract 1 Introduction 2 Related Work 2.1 Anchor-Based Method 2.2 Anchor-Free Me ...
＜Visual Relationship Detection with Internal and External Linguistic Knowledge Distillation＞论文阅读
论文链接:论文论文简介: 这是一篇CVPR2018的论文,主要针对的是Visual Relationship Detection任务.论文主要利用谓词及<object,subject>对 ...
论文阅读：Visual Relationship Detection with Language Priors
Visual Relationship Detection with Language Priors(ECCV2016) 文章尽管大多数的relationship并不常见,但是它们的object ...
论文阅读 End-to-End Multi-View Fusion for 3D Object Detection in Lidar Point Clouds
[论文阅读] End-to-End Multi-View Fusion for 3D Object Detection in Lidar Point Clouds 原文链接:https://arxiv ...
论文阅读Check it again:Progressive Visual Question Answering via Visual Entailment
论文:Check it again:Progressive Visual Question Answering via Visual Entailment 代码:https://github.com/ ...
【论文阅读】【综述】3D Object Detection 3D目标检测综述
目录写在开头 3D Object Detection 相关博客: Sliding window Vote3Deep: Fast Object Detection in 3D Point Clouds ...
【论文阅读】SCRDet：Towards More Robust Detection for Small, Cluttered and Rotated Objects
SCRDet: Towards More Robust Detection for Small, Cluttered and Rotated Objects SCRDet:为小的.杂乱的和旋转的物体提 ...
ICPR 2020 | 论文阅读 ——SyNet: An Ensemble Network for Object Detection in UAV Images
SyNet 1. Motivation 2. Method 2.1. Object detecion 2.2. SyNet 2.3 Image Augmentation 3. Experiments ...
[论文阅读]Contextual Instance Decoupling for Robust Multi-Person Pose Estimation
该论文发表于CVPR2022 Abstract 拥挤场景使得定位不同人体关键点具有挑战性.本文提出了一种上下文实例解耦(CID,Contextual Instance Decoupling)的新多人姿 ...

论文阅读：Contextual Translation Embedding for Visual Relationship Detection and SGG(PAMI2020)

论文阅读：Contextual Translation Embedding for Visual Relationship Detection and SGG(PAMI2020)相关推荐

最新文章

热门文章