【论文阅读】DetectoRS|Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution

题目：DetectoRS：Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution
作者：Siyuan Qiao Liang-Chieh Chen Alan Yuille

Motivation:
作者看到很多目标检测的网络都采用looking and thinking twice 的机制，然后作者也想来爽一把，于是乎牛逼的DetectoRS问世了。

Methods:
本文主要体现looking and thinking twice的思想，主要在两个方面：

1）RFP，Recursive Feature Pyramid

其中反向整合ASPP如下图：

最后FUSION的方式:

2）SAC，Switchable Atrous Convolution
可切换的空洞卷积比较容易理解，就像做了下个开关，然后选择是一般卷积还是空洞卷积。而作者具体的做法是采用AvgPool(5x5)+Conv(1x1)的形式做为选择方式，我觉得像是一个注意力机制的选择方式。

Experiment:
还是不多说，直接上图。效果很不错，谢谢作者的做贡献。

【完结】

【论文阅读】DetectoRS|Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution相关推荐

【论文笔记】：DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution
&Title DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolu ...
DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution 论文笔记
前言本文分别从宏观和微观角度对目标检测模型的backbone进行设计,进而提出DetectorRS, 在宏观上,提出递归特征金字塔(Recursive Feature Pyramid,RFP),它是 ...
DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution论文学习
Abstract 很多的目标检测器通过 looking and thinking twice 的方式实现了惊人的表现.本文作者针对目标检测主干网络的设计,研究了该机制.在宏观层面,作者提出了递归特征金 ...
论文导读：DetectoRS： Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution
论文导读:DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution A ...
DetectoRS-最强目标检测网络 54.7AP | Detecting Objects with Recursive Feature Pyramid and Switchable Atrous
最新目标检测论文,从检测的骨干网络设计出发(最近基于backbone的修改,貌似更有效!同比的由SCNet),效果直接SOTA! 论文地址:https://arxiv.org/pdf/2006.023 ...
ECCV2018 | 论文阅读CornerNet: Detecting Objects as Paired Keypoints
CornerNet论文阅读--CornerNet: Detecting Objects as Paired Keypoints 文章目录 CornerNet论文阅读--CornerNet: Detec ...
论文阅读 ACNet: Strengthening the Kernel Skeletons for Powerful CNN via Asymmetric Convolution Blocks
论文阅读 ACNet: Strengthening the Kernel Skeletons for Powerful CNN via Asymmetric Convolution Blocks 前言 ...
[HSI论文阅读] | HybridSN: Exploring 3-D–2-D CNN Feature Hierarchy for Hyperspectral Image Classification
仅整理个人和组里对下文引用的论文(HybridSN)的学习所获,如有错误,感谢指正~ 更新记录 ⭐️ 2021.3.6 -- 关于本文2.2.2节卷积层参数的计算 ⭐️ 2021.3.8 -- Hyb ...
【论文阅读】Detecting concurrency memory corruption vulnerabilities
Detecting concurrency memory corruption vulnerabilities 现状在多线程执行中会出现内存损坏的漏洞,在该篇论文称之为并发性漏洞.由于非确定性的多线 ...

【论文阅读】DetectoRS|Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution

【论文阅读】DetectoRS|Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution相关推荐

最新文章

热门文章