<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[PredragonTC的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/PredragonTC</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; PredragonTC]]></copyright><item><title><![CDATA[（还需要阅读代码）FoldingNet论文阅读]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/102240524</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/102240524</guid><author>PredragonTC</author><pubDate>Mon, 21 Oct 2019 16:50:24 +0800</pubDate><description><![CDATA[Main idea

]]></description><category></category></item><item><title><![CDATA[PointNetLK论文阅读]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/102650550</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/102650550</guid><author>PredragonTC</author><pubDate>Mon, 21 Oct 2019 16:50:09 +0800</pubDate><description><![CDATA[Main Idea

把点云和通过pointnet网络映射到和，目标是找出一个rigid transformation使得。借鉴LK的思想，LK目标是跟踪detector的平移，这里是跟踪整个点云的旋转。LK是有两个2D图片，，有提取出来的的detector，在中根据灰度找到的detector，算出两者位移。LK是基于光度不变假设，即，解出。这里是相似的，已知解出。这里是迭代求解之后，解出的与G...]]></description><category></category></item><item><title><![CDATA[3D Local Features for Direct Pairwise Registration论文阅读]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/101621594</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/101621594</guid><author>PredragonTC</author><pubDate>Mon, 21 Oct 2019 16:49:42 +0800</pubDate><description><![CDATA[Main Idea
Structure

]]></description><category></category></item><item><title><![CDATA[SDRSAC（CVPR2019）论文阅读]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/102639826</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/102639826</guid><author>PredragonTC</author><pubDate>Mon, 21 Oct 2019 16:49:28 +0800</pubDate><description><![CDATA[Main Idea

本文提出了一种不依靠descriptor计算出两个点云的相对，的方法。输入两片点云，分别从两片点云中随机抽取点数都为的两片小点云。运用纯数学的方法估计出correspondences，算出，。然后用ICP去refine这个，。经过多次随机抽取生成多个，，借鉴RANSAC，利用inlier ratio计算出终止条件，最终得出能够使最多的点匹配的那个，（全局重投影误差最小）

可...]]></description><category></category></item><item><title><![CDATA[C++中的cast_const]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/102569353</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/102569353</guid><author>PredragonTC</author><pubDate>Tue, 15 Oct 2019 16:22:12 +0800</pubDate><description><![CDATA[一. 函数描述：
const_cast &lt; type-id &gt; ( expression )
主要是用来去掉const属性，当然也可以加上const属性。主要是用前者，后者很少用。

去掉const属性：const_case&lt;int*&gt; (&amp;num)，常用，因为不能把一个const变量直接赋给一个非const变量，必须要转换。

加上const属性：const in...]]></description><category></category></item><item><title><![CDATA[L2-Net论文阅读]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/102295378</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/102295378</guid><author>PredragonTC</author><pubDate>Mon, 07 Oct 2019 12:26:48 +0800</pubDate><description><![CDATA[Main idea
这是把deep learning用于2D图像中进行patch match的research。

这个CS Network是这篇文章，主要思想是把一个6464的patch先downsample得到一个3232的低分辨率的patch1，然后在原先6464的patch的中心区域截取3232得到patch2，这两个patch分辨率一个高一个低，相当于中心区域的信息twice，还包含了整...]]></description><category></category></item><item><title><![CDATA[3Dmatch论文阅读]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/101054186</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/101054186</guid><author>PredragonTC</author><pubDate>Wed, 25 Sep 2019 16:38:05 +0800</pubDate><description><![CDATA[Main idea

]]></description><category></category></item><item><title><![CDATA[(还有疑问没解决)Multi-view Convolutional Neural Networks for 3D Shape Recognition论文阅读（ICCV2015）]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/98080916</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/98080916</guid><author>PredragonTC</author><pubDate>Thu, 15 Aug 2019 18:46:37 +0800</pubDate><description><![CDATA[MVCNN论文阅读InputRecognitionimage descriptorsclassificationretrieval measureAggregateLow-rank Mahalanobis metric如何插入一段漂亮的代码片
Input
这部分还需要结合代码来看，一会儿更新。
Recognition
image descriptors
使用VGG-M模型，主要结构为 conv[1...]]></description><category></category></item><item><title><![CDATA[Frustum PointNets for 3D Object Detection from RGB-D Data论文阅读(CVPR2018)]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/98217161</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/98217161</guid><author>PredragonTC</author><pubDate>Thu, 15 Aug 2019 18:44:55 +0800</pubDate><description><![CDATA[Problem Definition
3D detection with Frustum PointNets
Frustum Proposal
3D Instance Segmentation
3D Instance Segmentation PointNet
Amodal 3D Box Estimation
Learning-based 3D Alignment by T-Net
Amodal ...]]></description><category></category></item><item><title><![CDATA[leetcode总结]]></title><link>https://blog.youkuaiyun.com/PredragonTC/article/details/83656221</link><guid>https://blog.youkuaiyun.com/PredragonTC/article/details/83656221</guid><author>PredragonTC</author><pubDate>Fri, 02 Nov 2018 15:33:26 +0800</pubDate><description><![CDATA[融合有序数组or链表的一个思路


class Solution(object):
    def merge(self, nums1, m, nums2, n):
        &quot;&quot;&quot;
        :type nums1: List[int]
        :type m: int
        :type nums2: List[int]
        :type n: int
 ...]]></description><category></category></item></channel></rss>