【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*

论文

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第1张图片

论文题目:Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*

收录:CVPR2022

论文地址:[2203.14506] Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection (arxiv.org)

项目地址:GitHub - Choubo/DRA: Official PyTorch implementation of the paper “Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection”, open-set anomaly detection, few-shot anomaly detection.

论文翻译:Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection_appron的博客-CSDN博客

本文讲的是关于开放数据集上图像异常检测的问题,文中将很多知识框架信息都隐去了,想具体了解的可以先看另一篇论文《Deep Anomaly Detection with Deviation Networks》,几篇解读写的很明白,看完后就基本能理解本文的框架了。

《Deep Anomaly Detection with Deviation Networks》

论文来源 | KDD 2019
论文链接 | [1911.08623] Deep Anomaly Detection with Deviation Networks (arxiv.org)
源码链接 | GitHub - GuansongPang/deviation-network: Source code of the KDD19 paper "Deep anomaly detection with deviation networks", weakly/partially supervised anomaly detection, few-shot anomaly detection

论文解读 | DevNet:基于偏差网络的深度异常检测模型 | 梦家博客 (dreamhomes.top)

论文分享 | Deep Anomaly Detection with Deviation Networks (qq.com)

DevNet半监督异常识别模型 - 知乎 (zhihu.com)

整体框架图解读:Deep Anomaly Detection with Deviation Networks

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第2张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第3张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第4张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第5张图片

 

论文解读 Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*

数据集展示 

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第6张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第7张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第8张图片

  

论文改进

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第9张图片

主要内容&贡献

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第10张图片

 关键点

【读论文04】CVPR2022选读_要谦年人的博客-CSDN博客

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第11张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第12张图片

问题定义

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第13张图片

 思路

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第14张图片

框架

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第15张图片

代码解读

4个head 

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第16张图片

 数据预处理

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第17张图片

可见异常样本读取 

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第18张图片

 伪异常样本生成

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第19张图片

 模型输入

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第20张图片

解耦异常得分 

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第21张图片

返回值

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第22张图片

Loss 

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第23张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第24张图片

BEC Loss、Focal Loss和Dev Loss的比较 

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第25张图片

分离异常学习,各head的loss 

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第26张图片

实验

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第27张图片

实验细节 

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第28张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第29张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第30张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第31张图片

普通设定结果

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第32张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第33张图片

困难设定结果

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第34张图片

消融实验

每个异常学习head的重要性

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第35张图片

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第36张图片

伪异常样本生成方式比较

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第37张图片

解耦学习的重要性 及 参考图像数量 比较 

【论文笔记】Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection*_第38张图片

你可能感兴趣的:(论文笔记,深度学习,人工智能)