<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[Orientliu96的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/Orientliu96</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; Orientliu96]]></copyright><item><title><![CDATA[[Open3D WARNING] GLFW Error: X11: Failed to open display :0]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/131840707</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/131840707</guid><author>Orientliu96</author><pubDate>Thu, 20 Jul 2023 21:52:27 +0800</pubDate><description><![CDATA[linux 系统无法打开open3d的可视化窗口，报错信息如下。]]></description><category></category></item><item><title><![CDATA[Latex公式炫酷技巧]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/131643225</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/131643225</guid><author>Orientliu96</author><pubDate>Mon, 10 Jul 2023 17:01:31 +0800</pubDate><description><![CDATA[最近看到一个炫酷的latex公式用法，特意在此记录一下。]]></description><category></category></item><item><title><![CDATA[pytorch grid_sample易错点]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/131642445</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/131642445</guid><author>Orientliu96</author><pubDate>Mon, 10 Jul 2023 16:51:17 +0800</pubDate><description><![CDATA[output的size和grid的size是一样的，所以output中某一位置(h, w)的值，是根据grid的同一位置(h,w)提供的index信息去input中寻找。如果input的网格中存在这个索引，直接赋值。若不存在，则进行插值计算。那么如果想得到非转置的结果，就要将grid更改一下。output是input的转置。]]></description><category></category></item><item><title><![CDATA[Specify CUDA_TOOLKIT_ROOT_DIR]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/111884804</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/111884804</guid><author>Orientliu96</author><pubDate>Tue, 29 Dec 2020 11:08:52 +0800</pubDate><description><![CDATA[Specify CUDA_TOOLKIT_ROOT_DIR

出现这个错误的原因是编译器没有找到cuda的路径，所以需要我们告诉他应该去哪里找我们的cuda文件
cmake -D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-8.0 ..

手动添加路径后，编译成功。


...]]></description><category></category></item><item><title><![CDATA[逐行统计python代码执行时间]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/111305002</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/111305002</guid><author>Orientliu96</author><pubDate>Wed, 16 Dec 2020 23:09:14 +0800</pubDate><description><![CDATA[逐行统计python代码执行时间
在炼丹或者debug的过程中，我们经常需要去逐行统计代码运行的时间。使用time.time()是可以解决问题，但是过于繁琐。于是通过一顿踩坑，推荐使用line_profiler库, 安装方式如下，十分简单。
pip install line_profiler


安装成功，下面开始介绍如何使用。


from __future__ import print_function

from line_profiler import LineProfiler
import tor]]></description><category></category></item><item><title><![CDATA[Open3D 可视化numpy点云数据]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/110948330</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/110948330</guid><author>Orientliu96</author><pubDate>Thu, 10 Dec 2020 11:00:21 +0800</pubDate><description><![CDATA[Open3D 可视化numpy点云数据
使用Open3D 可视化点云数据的效果比较炫。
但是当我时隔半年重新使用Open3D库时发现原来的用法发生了改变，特此记录，持续更新。
Anaconda 安装Open3D
官方网站
conda install -c open3d-admin open3d

比较稳定，推荐此方式
导入numpy点云数据，并可视化
import numpy as np
from open3d import*

source_data = np.load('curtain_0088.npy]]></description><category></category></item><item><title><![CDATA[LaTeX 取消algorithm中的行编号]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/110809544</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/110809544</guid><author>Orientliu96</author><pubDate>Mon, 07 Dec 2020 12:27:43 +0800</pubDate><description><![CDATA[LaTeX Error: File `geometry.sty’ not found
将导入包命令
\usepackage[linesnumbered,lined,boxed]{algorithm2e}

改为
\usepackage[lined,boxed]{algorithm2e}

就可以达到如下效果


]]></description><category></category></item><item><title><![CDATA[【点云识别】A Closer Look at Local Aggregation Operators in Point Cloud Analysis (ECCV 2020)]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/109569627</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/109569627</guid><author>Orientliu96</author><pubDate>Mon, 09 Nov 2020 10:34:18 +0800</pubDate><description><![CDATA[A Closer Look at Local Aggregation Operators in Point Cloud Analysis
本文介绍一篇ECCV2020里面关于点云局部聚合操作的文章。
论文
代码
1. 问题
点云的局部聚合操作
2. 思想

不需要学习参数，简洁高效。
3. Overview of Local Aggregation Operators
General Formulation

3.1 Point-wise MLP based Methods

3.2 Pseudo Grid]]></description><category></category></item><item><title><![CDATA[【点云识别】Modeling Local Geometric Structure of 3D Point Clouds using Geo-CNN (CVPR 2019)]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/108724899</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/108724899</guid><author>Orientliu96</author><pubDate>Tue, 22 Sep 2020 10:01:25 +0800</pubDate><description><![CDATA[Modeling Local Geometric Structure of 3D Point Clouds using Geo-CNN
本文介绍一篇cvpr2019里面关于点云处理的文章。
论文
代码
1. 问题
点云处理的方式
2. 思想

3. 算法

利用坐标轴，建立点云特征的相互关系。详见论文作者的解析
4 实验结果

分类效果还是很棒的。

总结
文章思考的角度很新颖，分类实验的效果也很棒。唯一的疑问是，这种稍显复杂的特征提取方式的效率如何？如果能在训练时间、inference time上和poi]]></description><category></category></item><item><title><![CDATA[【点云识别】PointContrast: Unsupervised Pre-training for 3D Point Cloud Understanding (ECCV 2020)]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/108007214</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/108007214</guid><author>Orientliu96</author><pubDate>Fri, 14 Aug 2020 17:00:20 +0800</pubDate><description><![CDATA[PointContrast: Unsupervised Pre-training for 3D Point Cloud Understanding
本文介绍一篇ECCV2020里面关于点云无监督预训练的文章。
论文
目前还没有开源代码
1. 问题
点云无监督预训练
2. 思想

点云的预训练领域处于比较空白的状态，本文focus on 高级别的场景理解任务，提出了一种无监督的PointContrast与训练方法。文章主体依赖于FCGF, 以及无监督学习 3D representation的思想（throug]]></description><category></category></item><item><title><![CDATA[LaTeX引用表格的编号错误]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/107752537</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/107752537</guid><author>Orientliu96</author><pubDate>Sun, 02 Aug 2020 22:13:31 +0800</pubDate><description><![CDATA[LaTeX引用表格的编号错误
部分期刊模板下表格的引用标号会出现错误，例如我引用第一个表格会出现

解决办法：
lable要紧跟在caption后面
\caption{1111111111111111111111111111.}
\label{1}


]]></description><category></category></item><item><title><![CDATA[【点云识别】Mapping in a Cycle: Sinkhorn Regularized Unsupervised Learning for Point Cloud Shapes ECCV2020]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/107610705</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/107610705</guid><author>Orientliu96</author><pubDate>Mon, 27 Jul 2020 15:30:20 +0800</pubDate><description><![CDATA[Mapping in a Cycle: Sinkhorn Regularized Unsupervised Learning for Point Cloud Shapes
本文介绍一篇ECCV2020里面关于点云无监督学习的文章。
论文
目前还没有开源代码
1. 问题
点云形状的无监督学习
2. 思想

本文的主体思路是延续 Unsupervised cycle-consistent deformation for shape matching 这篇文章，在其基础上添加了sinkhorn regulariz]]></description><category></category></item><item><title><![CDATA[ImportError: /lib64/libm.so.6: version `GLIBC_2.27‘ not found]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/107048827</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/107048827</guid><author>Orientliu96</author><pubDate>Tue, 30 Jun 2020 19:27:51 +0800</pubDate><description><![CDATA[ImportError: /lib64/libm.so.6: version `GLIBC_2.27’ not found (required by /home/labpos/anaconda3/envs/rpmnet/lib/python3.6/site-packages/open3d/open3d_pybind.cpython-36m-x86_64-linux-gnu.so)

这应该是open3d版本不兼容引起的，将版本卸载，重新安装0.9版本的open3d。
conda install -c ope]]></description><category></category></item><item><title><![CDATA[【点云识别】Feature-metric Registration: A Fast Semi-supervised Approach for Robust Point Cloud(CVPR 2020)]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/106888829</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/106888829</guid><author>Orientliu96</author><pubDate>Sun, 21 Jun 2020 18:13:47 +0800</pubDate><description><![CDATA[Feature-metric Registration: A Fast Semi-supervised Approach for Robust Point Cloud Registration without Correspondences
本文介绍一篇cvpr2020里面关于点云半监督配准的文章。
论文
没有开源代码
1. 问题
点云的弱监督配准问题
2. 思想

follow pointlk的一份工作，认为不同位姿的点云提取出的特征是不同的。那么将point set的global feature res]]></description><category></category></item><item><title><![CDATA[【点云识别】Adaptive Hierarchical Down-Sampling for Point Cloud Classification(CVPR 2020)]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/106762832</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/106762832</guid><author>Orientliu96</author><pubDate>Mon, 15 Jun 2020 15:40:23 +0800</pubDate><description><![CDATA[Adaptive Hierarchical Down-Sampling for Point Cloud Classification
本文介绍一篇cvpr2020里面关于点云分类降采样的文章。
论文
没有开源代码
1. 问题
FPS的时间复杂度太高，类似Samplenet 的方法会产生新的点，随机采样无法保证重要的点被保留下来。所以本文提出了一种不产生新的点的采样方法。
2. 思想

整体思想非常简单，借鉴pointnet中最后的maxpooling操作。将每个feature属性最max的点视为critic]]></description><category></category></item><item><title><![CDATA[【点云识别】D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features(CVPR 2020 Oral)]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/106542947</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/106542947</guid><author>Orientliu96</author><pubDate>Thu, 04 Jun 2020 15:01:06 +0800</pubDate><description><![CDATA[D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features
本文介绍一篇cvpr2020里面关于点云matching的文章。
论文
代码
1. 问题
Point Cloud Matching。 前人的工作没有将点云的dectection和descriptors联合起来学习， 会导致detector 不能匹配上 descriptor的能力。
2. 思想
同时进行dectection and description]]></description><category></category></item><item><title><![CDATA[【点云识别】Unsupervised Learning of Intrinsic Structural Representation Points(CVPR 2020)]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/106364158</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/106364158</guid><author>Orientliu96</author><pubDate>Tue, 26 May 2020 21:36:42 +0800</pubDate><description><![CDATA[Unsupervised Learning of Intrinsic Structural Representation Points
本文介绍一篇cvpr2020里面无监督学习内在结构表征点的文章。
论文
代码
1. 问题
无监督学习内在结构表征点
2. 思想

用chamfer loss约束重建点云的质量，利用point integration module 来寻找 structure points.
3. 算法

给人感觉更像是key point detection,包装成structure poin]]></description><category></category></item><item><title><![CDATA[Package inputenc Error: Invalid UTF-8 byte sequence. [This electronic document is a l]]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/106276373</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/106276373</guid><author>Orientliu96</author><pubDate>Fri, 22 May 2020 11:32:22 +0800</pubDate><description><![CDATA[Package inputenc Error: Invalid UTF-8 byte sequence. [This electronic document is a l]

问题很好解决，在文章的开头添加
\UseRawInputEncoding


就可以解决问题了

]]></description><category></category></item><item><title><![CDATA[LaTeX Error: File `.def’ not found. [\usepackage]]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/106207202</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/106207202</guid><author>Orientliu96</author><pubDate>Tue, 19 May 2020 09:20:48 +0800</pubDate><description><![CDATA[LaTeX Error: File `.def’ not found. [\usepackage]

出现这个问题上是因为我的电脑中及安装了Ctex和latex,所以会产生冲突。卸载Ctex即可。

]]></description><category></category></item><item><title><![CDATA[【点云识别】Weakly Supervised Semantic Point Cloud Segmentation: Towards 10x Fewer Labels(CVPR 2020)]]></title><link>https://blog.youkuaiyun.com/Orientliu96/article/details/105703067</link><guid>https://blog.youkuaiyun.com/Orientliu96/article/details/105703067</guid><author>Orientliu96</author><pubDate>Thu, 23 Apr 2020 12:04:39 +0800</pubDate><description><![CDATA[Weakly Supervised Semantic Point Cloud Segmentation: Towards 10x Fewer Labels
本文介绍一篇cvpr2020里面关于点云弱监督分割的文章。
论文
目前还没有开源代码
1. 问题
点云的弱监督分割
2. 思想

根据弱监督的特点，提出了incomplete supervision branch 和 inexact super...]]></description><category></category></item></channel></rss>