<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[s000da的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/s000da</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; s000da]]></copyright><item><title><![CDATA[AI提效工具｜借助chatgpt快速读论文，快速总结、归纳、索引相似文章]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/130077670</link><guid>https://blog.youkuaiyun.com/s000da/article/details/130077670</guid><author>s000da</author><pubDate>Tue, 11 Apr 2023 10:47:14 +0800</pubDate><description><![CDATA[本文展示了两个借助chatgpt快速总结论文的工具，让你读论文速度快到飞起～]]></description><category></category></item><item><title><![CDATA[半监督-SelfMatch-论文阅读笔记]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/117263271</link><guid>https://blog.youkuaiyun.com/s000da/article/details/117263271</guid><author>s000da</author><pubDate>Tue, 25 May 2021 19:56:38 +0800</pubDate><description><![CDATA[阅读背景

SimCLR在2020年2月第一次挂在arxiv上，被ICML 2020接收

FixMatch在2020年1月第一次挂在arxiv上，被NIPS 2020接收

概括总结

SelfMatch方法和 FixMatch 都是NIPS 2020的文章，前者是work shop，后者是conference。

核心方法，是将自监督的contrastive learning 和 半监督的 consistency regularization 结合，具体是SimCLR+FixMatch 的流程：首先在u]]></description><category></category></item><item><title><![CDATA[动手实践——docker中利用jupyter对数据增强操作进行可视化]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/115013548</link><guid>https://blog.youkuaiyun.com/s000da/article/details/115013548</guid><author>s000da</author><pubDate>Fri, 19 Mar 2021 19:03:50 +0800</pubDate><description><![CDATA[整体流程

docker容器内搭建合适环境 -&gt; 开启jupyter notebook -&gt; 浏览器里敲数据增强操作代码 -&gt;可视化

搭建环境

参考博客https://segmentfault.com/a/1190000007448177

1. 启动容器

2. 安装openssh-server：


apt-get install openssh-server

3. 编辑配置文件/etc/ssh/sshd_config，注释掉配置文件中的"PermitRootLogin ...]]></description><category></category></item><item><title><![CDATA[半监督领域论文笔记——Billion-scale semi-supervised learning for image classification]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/109232063</link><guid>https://blog.youkuaiyun.com/s000da/article/details/109232063</guid><author>s000da</author><pubDate>Thu, 22 Oct 2020 23:05:41 +0800</pubDate><description><![CDATA[发表时间

2019年

所属领域

半监督学习

文章方法

主要目的

借助非标注数据，提升现有模型效果

方法概述

采用了teacher/student的学习机制，借助了billion级别的unable data和相对小数量级的label data，提升了当前已有模型在图像分类任务上的效果

提出背景

2018年，也是Facebook，提出了weakly supervised方向的研究“Exploring the Limits of Weakly Supervised Pretraining”，采]]></description><category></category></item><item><title><![CDATA[论文笔记整理模板]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/109094581</link><guid>https://blog.youkuaiyun.com/s000da/article/details/109094581</guid><author>s000da</author><pubDate>Thu, 15 Oct 2020 11:56:19 +0800</pubDate><description><![CDATA[为了看论文时更有条理，并且沉淀更多，总结了一个自用的【论文笔记整理模板】，以xmind格式呈现。



免费下载链接
]]></description><category></category></item><item><title><![CDATA[半监督学习方向——SimCLR论文阅读笔记]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/108585993</link><guid>https://blog.youkuaiyun.com/s000da/article/details/108585993</guid><author>s000da</author><pubDate>Mon, 14 Sep 2020 19:28:49 +0800</pubDate><description><![CDATA[参考资料：

https://medium.com/%E8%BB%9F%E9%AB%94%E4%B9%8B%E5%BF%83/deep-learning-self-supervised-learning%E7%9A%84%E6%96%B0%E9%AB%98%E5%BA%A6-%E5%BE%9Esimclr%E8%AA%8D%E8%AD%98ssl%E8%88%87contrastive-learning-5dad7a7097e1
]]></description><category></category></item><item><title><![CDATA[PyTorch模型结构可视化方法总结（未完结）]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/106074709</link><guid>https://blog.youkuaiyun.com/s000da/article/details/106074709</guid><author>s000da</author><pubDate>Tue, 12 May 2020 15:50:42 +0800</pubDate><description><![CDATA[总体来说有几种方法：netron、

1. netron（个人最喜欢）

具体方法：pytorch存下来的.pth文件转为.onnx文件，直接拖入netron网页就可以看
	优点：画出来比较细致，布局效果好，直观
2. tensorboard

缺点：加上tensorboard中关于summarywriter的代码后，好像会让显存增加不少，而且第一张卡的显存特别多
...]]></description><category></category></item><item><title><![CDATA[ps篡改方向Learning Rich Features for Image Manipulation Detection——论文笔记（未完成）]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/105049786</link><guid>https://blog.youkuaiyun.com/s000da/article/details/105049786</guid><author>s000da</author><pubDate>Mon, 23 Mar 2020 15:37:25 +0800</pubDate><description><![CDATA[论文链接





拓展阅读

知乎，论文笔记-篡改方向






问题记录

怎样从RGB图像通过SRM filter kernels获取noise feature？



]]></description><category></category></item><item><title><![CDATA[label smooth方法论文调研]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/104802160</link><guid>https://blog.youkuaiyun.com/s000da/article/details/104802160</guid><author>s000da</author><pubDate>Wed, 11 Mar 2020 18:09:20 +0800</pubDate><description><![CDATA[待看论文：


	When Does Label Smoothing Help?（重点要看的）
	
	
	Regularizing Neural Networks by Penalizing Confident Output Distributions
	
	
	On Calibration of Modern Neural Networks （关于模型calibration）
	
...]]></description><category></category></item><item><title><![CDATA[learning with noisy labels方向论文笔记（待更新）]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/103351775</link><guid>https://blog.youkuaiyun.com/s000da/article/details/103351775</guid><author>s000da</author><pubDate>Tue, 03 Dec 2019 15:38:37 +0800</pubDate><description><![CDATA[博客笔记：

1. 几篇learning with noisy labels论文的总结

博客地址：https://blog.youkuaiyun.com/ferriswym/article/details/89452430#t3

笔记：



相关论文阅读计划：

Probabilistic End-to-end Noise Correction for Learning with Noisy Labe...]]></description><category></category></item><item><title><![CDATA[图像去重工具——imagededup（未完待更新）]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/103299574</link><guid>https://blog.youkuaiyun.com/s000da/article/details/103299574</guid><author>s000da</author><pubDate>Thu, 28 Nov 2019 19:51:23 +0800</pubDate><description><![CDATA[去重原理





工具介绍

链接：https://github.com/idealo/imagededup



使用经验

在无脑搬运并使用此黑箱时，本小白耗时快一天，遇到了以下几方面的问题：

配置安装环境：python版本不适合，pip安装imagededup出错
	调相似度阈值
接下来记录以上环节的具体解决情况。


	配置安装环境
	
使用pip安装时制定python版本，并且手...]]></description><category></category></item><item><title><![CDATA[问题记录——PIL读图、opencv读图、mxnet读图后存在差异]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/103032783</link><guid>https://blog.youkuaiyun.com/s000da/article/details/103032783</guid><author>s000da</author><pubDate>Tue, 12 Nov 2019 16:50:47 +0800</pubDate><description><![CDATA[遇到问题

同一张图片，经PIL的Image.open读取，和经cv2.imread读取，两种方式得到的图像像素值存在差异。同时，cv2.imread得到的结果和mxnet.image.imread是一样的（但也遇到了某些神图，mxnet.image.imread读进来完全是噪声，而cv2读进来就是正常的情况）。分析原因是因为mxnet.image.imread的底层实现也是的opencv。

总...]]></description><category></category></item><item><title><![CDATA[深浅层特征融合——HRNet]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/102516264</link><guid>https://blog.youkuaiyun.com/s000da/article/details/102516264</guid><author>s000da</author><pubDate>Sat, 12 Oct 2019 11:05:17 +0800</pubDate><description><![CDATA[写在前面

本系列博客“深浅层特征融合”对几篇出现较新的深浅层特征融合算法进行简要介绍，多为本人的论文笔记，记录了一个深度学习小学生在看论文时想到的问题。

论文题目：Deep High-Resolution Representation Learning for Human Pose Estimation

论文链接：https://arxiv.org/pdf/1902.09212.pdf

g...]]></description><category></category></item><item><title><![CDATA[深浅层特征融合——CBNet]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/102495156</link><guid>https://blog.youkuaiyun.com/s000da/article/details/102495156</guid><author>s000da</author><pubDate>Fri, 11 Oct 2019 10:02:14 +0800</pubDate><description><![CDATA[写在前面

本系列博客“深浅层特征融合”对几篇出现较新的深浅层特征融合算法进行简要介绍，多为本人的论文笔记，记录了一个深度学习小学生在看论文时想到的问题。

论文题目：CBNet: A Novel Composite Backbone Network Architecture for Object Detection

论文链接：https://arxiv.org/pdf/1909.03625.p...]]></description><category></category></item><item><title><![CDATA[学习计划]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/102465668</link><guid>https://blog.youkuaiyun.com/s000da/article/details/102465668</guid><author>s000da</author><pubDate>Wed, 09 Oct 2019 17:01:50 +0800</pubDate><description><![CDATA[总体方向

更合适的特征表达（借鉴目标检测领域算法）
	attention（可以从最经典的SENet开始尝试）
	半监督

]]></description><category></category></item><item><title><![CDATA[pytorch自学入门记录]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/101376697</link><guid>https://blog.youkuaiyun.com/s000da/article/details/101376697</guid><author>s000da</author><pubDate>Wed, 25 Sep 2019 17:46:31 +0800</pubDate><description><![CDATA[前提：

接触过Keras、mxnet等深度学习框架
自学资料：

根据知乎回答，参考别人学习的顺序


官方tutorial（60min），英文，中文翻译版本
	比较靠谱的介绍博客：https://blog.youkuaiyun.com/u014380165/article/category/9271909
...]]></description><category></category></item><item><title><![CDATA[今天又犯了哪些错]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/99960287</link><guid>https://blog.youkuaiyun.com/s000da/article/details/99960287</guid><author>s000da</author><pubDate>Wed, 21 Aug 2019 14:30:50 +0800</pubDate><description><![CDATA[RT，本问记录下来每天犯的错误。。

2019.08.20

用多个模型的阈值来卡数据，获取有用数据进行标注加入训练集。卡阈值的时候不应指定模型，这样会卡得很难，可以卡模型个数。


...]]></description><category></category></item><item><title><![CDATA[图像篡改检测方向（Tamper Detection\ image forencis detection）学习笔记汇总（未完成）]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/95206073</link><guid>https://blog.youkuaiyun.com/s000da/article/details/95206073</guid><author>s000da</author><pubDate>Tue, 09 Jul 2019 17:16:08 +0800</pubDate><description><![CDATA[参考资料：

关于图像篡改检测的算法总结：

http://homeway.me/2018/08/05/tamper-detection-an-overview/
	http://homeway.me/2019/05/04/tamper-detection-an-overview-2/
	https://blog.youkuaiyun.com/u010183397/article/details/547994...]]></description><category></category></item><item><title><![CDATA[深度学习——不同的卷积形式]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/94334744</link><guid>https://blog.youkuaiyun.com/s000da/article/details/94334744</guid><author>s000da</author><pubDate>Sun, 30 Jun 2019 16:48:22 +0800</pubDate><description><![CDATA[写在前面

本文总结了不同的卷积形式以及各自的参数量、计算量。

具体总结

分组卷积（Group convolution）



空洞卷积（Dilated convolution 或 Atrous convolution）



深度可分离卷积（depthwise separable convolutions）



参考博客

https://blog.youkuaiyun.com/u0104171...]]></description><category></category></item><item><title><![CDATA[深度学习论文阅读计划（超级天坑，待填）]]></title><link>https://blog.youkuaiyun.com/s000da/article/details/93617223</link><guid>https://blog.youkuaiyun.com/s000da/article/details/93617223</guid><author>s000da</author><pubDate>Tue, 25 Jun 2019 16:32:23 +0800</pubDate><description><![CDATA[写在前面

本文总结了自己在深度学习领域读过的一点论文，链接了自己或他人写的一些论文笔记也包含了对将来制定的论文阅读计划。将阅读笔记整理在一起是为了能融会贯通，力求梳理出统一解决某一类问题的技巧或思想，也为了成全自己在成长道路上的“满足感”。欢迎沟通， 与君共勉。

Table of Contents

写在前面

简要汇总

trick

轻量网络

CNN经典结构

目标检测算法

人脸识别...]]></description><category></category></item></channel></rss>