<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[sinat_35177634的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/sinat_35177634</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; sinat_35177634]]></copyright><item><title><![CDATA[因果推理（causal inference）初步调研]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/106097910</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/106097910</guid><author>sinat_35177634</author><pubDate>Wed, 13 May 2020 15:41:21 +0800</pubDate><description><![CDATA[因果推理（causal inference）初步调研
NIPS相关论文
NIPS有相关的causal inference板块，但总体来看，相比于较为成型的visual板块显得更加五花八门，少有一个统一的框架来解决关于因果推理的任务，甚至对任务的界定也比较模糊，大家各自为战。
比如treatment effects estimation任务，就包括individual treatment effects estimation和Heterogeneous Treatment Effects estimation]]></description><category></category></item><item><title><![CDATA[ICCV 2019 image-text相关论文总结]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/103257174</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/103257174</guid><author>sinat_35177634</author><pubDate>Tue, 26 Nov 2019 16:49:05 +0800</pubDate><description><![CDATA[ICCV 2019 image-text相关论文总结
1. Local Relation Networks for Image Recognition
issue: CNN用固定的filters来提取image feature，但image的空间分布是多变的，不能很好的满足image不同的空间分布。
method: 提出local relation layer，用于改进CNN。通过局部区域的关系(...]]></description><category></category></item><item><title><![CDATA[ICCV 2019 图像描述(image caption)论文汇总]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/103237609</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/103237609</guid><author>sinat_35177634</author><pubDate>Tue, 26 Nov 2019 14:58:12 +0800</pubDate><description><![CDATA[ICCV 2019 图像描述(image caption)论文汇总
1. Attention on Attention for Image Captioning
issue: decoder不能很好的分辨attended vector和当前生成语义的关系

method: 再使用一个卷积attention机制，输入为attended vector和当前语义信息，生成attention gate和i...]]></description><category></category></item><item><title><![CDATA[计算机算法设计与分析解题心得]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/100986463</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/100986463</guid><author>sinat_35177634</author><pubDate>Wed, 18 Sep 2019 16:04:00 +0800</pubDate><description><![CDATA[计算机算法设计与分析
Session 1

先观察问题的结构、解的形式，再设计算法；
能分解成子问题，是非常有效的信息；
优化问题时，下界很重要。

Session 2

给一个问题，从最简单的case入手；
观察INPUT的关键结构，看能否分、能否combine；
观察OUTPUT，不要坚持追求optimal，追求sub-optimal。

Session 3
Sample.
Session 4...]]></description><category></category></item><item><title><![CDATA[2018 NIPS LinkNet: Relational Embedding for Scene Graph论文解读]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/100898757</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/100898757</guid><author>sinat_35177634</author><pubDate>Tue, 17 Sep 2019 21:26:46 +0800</pubDate><description><![CDATA[2018 NIPS LinkNet: Relational Embedding for Scene Graph论文解读

]]></description><category></category></item><item><title><![CDATA[Leetcode C++ 33. Search in Rotated Sorted Array]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/100747236</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/100747236</guid><author>sinat_35177634</author><pubDate>Wed, 11 Sep 2019 20:07:44 +0800</pubDate><description><![CDATA[Leetcode 33. Search in Rotated Sorted Array
原文：
Suppose an array sorted in ascending order is rotated at some pivot unknown to you beforehand.
(i.e., [0,1,2,4,5,6,7] might become [4,5,6,7,0,1,2]).
You...]]></description><category></category></item><item><title><![CDATA[Leetcode C++ 31. Next Permutation]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/100711927</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/100711927</guid><author>sinat_35177634</author><pubDate>Tue, 10 Sep 2019 21:14:42 +0800</pubDate><description><![CDATA[Leetcode 31. Next Permutation
解题思路：输出字典序中的下一个排列。比如123生成的全排列是：123，132，213，231，312，321。那么321的next permutation是123。下面这种算法据说是STL中的经典算法。在当前序列中，从尾端往前寻找两个相邻升序元素，升序元素对中的前一个标记为k - 1。然后再从尾端寻找另一个大于k - 1的元素，并与k -...]]></description><category></category></item><item><title><![CDATA[CVPR2019 图像描述（image  caption）论文汇总]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/100666419</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/100666419</guid><author>sinat_35177634</author><pubDate>Mon, 09 Sep 2019 16:48:12 +0800</pubDate><description><![CDATA[CVPR2019 图像描述（image  caption）论文汇总
1.Adversarial Semantic Alignment for Improved Image Captions
使用对抗学习的思想来加强image和caption之间的关联
2.Context and Attribute Grounded Dense Captioning
对图像的密集描述，对每一个区域生成短语或者一个句...]]></description><category></category></item><item><title><![CDATA[图像描述（image caption）2019 CVPR oral论文总结]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/100269915</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/100269915</guid><author>sinat_35177634</author><pubDate>Tue, 03 Sep 2019 14:01:08 +0800</pubDate><description><![CDATA[图像描述（image caption）2019 CVPR oral论文总结
1. Auto-Encoding Scene Graphs for Image Captioning
论文链接：https://arxiv.org/pdf/1812.02378.pdf
2. Fast, Diverse and Accurate Image Captioning Guided By Part-of-Spee...]]></description><category></category></item><item><title><![CDATA[图像描述（image caption）2018 NIPS论文总结]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/100197780</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/100197780</guid><author>sinat_35177634</author><pubDate>Mon, 02 Sep 2019 15:42:23 +0800</pubDate><description><![CDATA[图像描述（image caption）2018 NIPS论文总结
1. Partially-Supervised Image Captioning
这篇文章是半监督来训练image caption的任务，因为image caption的数据集相对图像分类、目标检测等的数据集小的多，限制了image caption模型的泛化应用性。因此，本文考虑不使用传统的image caption数据集，仅仅使用...]]></description><category></category></item><item><title><![CDATA[CVPR 2019 Unsupervised Image Caption]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/100164699</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/100164699</guid><author>sinat_35177634</author><pubDate>Fri, 30 Aug 2019 20:36:52 +0800</pubDate><description><![CDATA[CVPR 2019 Unsupervised Image Caption
无监督的图像描述，和之前的全监督图像描述相比，它不使用图像和描述之间的关联关系，而仅仅使用图像集、一个额外的语料库和一个目标检测器来实现在这个图像集上的描述。使用的图像集是COCO2014，而使用的额外语料库是作者从ShutterStock上扒下来的描述集。
由于没有图像和描述之间的关联关系，作者给生成的描述设定了三个目标：...]]></description><category></category></item><item><title><![CDATA[CVPR 2019视频描述（video caption）相关论文总结]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/89601950</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/89601950</guid><author>sinat_35177634</author><pubDate>Sat, 27 Apr 2019 15:07:02 +0800</pubDate><description><![CDATA[CVPR 2019视频描述（video caption）相关论文总结
前几年的video caption论文总结如下：
https://blog.youkuaiyun.com/sinat_35177634/article/details/88568491
Streamlined Dense Video Captioning
这篇论文严格来讲是属于video caption的一个分支dense video ca...]]></description><category></category></item><item><title><![CDATA[temporal action proposals论文总结]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/88918421</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/88918421</guid><author>sinat_35177634</author><pubDate>Sun, 31 Mar 2019 11:07:03 +0800</pubDate><description><![CDATA[temporal action proposals论文总结——DAPs与SST
temporal action proposals
其主要目的是将长视频根据语义分割成多个segment，因为现在的针对视频的任务对长视频处理并不理想，比如视频的action detection和caption等。因此需要现将长视频分割成多个短视频，再进行处理。temporal action proposals是根据长...]]></description><category></category></item><item><title><![CDATA[密集事件描述(dense event caption)论文总结]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/88782635</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/88782635</guid><author>sinat_35177634</author><pubDate>Sat, 30 Mar 2019 20:28:01 +0800</pubDate><description><![CDATA[多事件描述(dense event caption)历年突破性论文总结
多事件描述（DEC）
何为多事件描述（DEC），其实它是视频描述（video caption）的一个分支，视频描述自2015年的S2VT方法以来开始广泛的使用深度神经网络尤其是RNN模型来解决视频描述任务，本人也写了本文也写了视频描述论文相关的综述：
https://blog.youkuaiyun.com/sinat_35177634/a...]]></description><category></category></item><item><title><![CDATA[视频描述（video caption）历年突破性论文总结]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/88568491</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/88568491</guid><author>sinat_35177634</author><pubDate>Fri, 15 Mar 2019 15:59:04 +0800</pubDate><description><![CDATA[视频描述（video caption）历年突破性论文总结（2015—2016）
视频描述
顾名思义视频描述是计算机对视频生成一段描述，如图所示，这张图片选取了一段视频的两帧，针对它的描述是&amp;quot;A man is doing stunts on his bike&amp;quot;，这对在线的视频的检索等有很大帮助。近几年图像描述的发展也让人们思考对视频生成描述，但不同于图像这种静态的空间信息，视频除了空间信息还包括时序...]]></description><category></category></item><item><title><![CDATA[图像描述（image caption）历年突破性论文总结]]></title><link>https://blog.youkuaiyun.com/sinat_35177634/article/details/88102512</link><guid>https://blog.youkuaiyun.com/sinat_35177634/article/details/88102512</guid><author>sinat_35177634</author><pubDate>Tue, 05 Mar 2019 11:50:41 +0800</pubDate><description><![CDATA[图像描述（image caption）历年突破性论文总结
图像描述(image caption)
顾名思义，图像描述是指以图像为输出，通过模型和计算来输出对应图像的自然语言描述。如下图所示，输出的自然语言是“A person riding a motorcycle on a dirt road.“ 。这一领域是结合了人工智能两大方向：计算机视觉和自然语言处理。从2015年的论文：Show and ...]]></description><category></category></item></channel></rss>