<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[Sun7_She的专栏]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/Sun7_She</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; Sun7_She]]></copyright><item><title><![CDATA[torch7安装教程]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/126732187</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/126732187</guid><author>Sun7_She</author><pubDate>Tue, 06 Sep 2022 19:35:40 +0800</pubDate><description><![CDATA[在装zeromq之前还要保证有autoconf和automake，安装语句为下边代码块的前两句。1. 其中安装JUPYTER的时候使用下边这句话，否则在国内似乎不能访问文中的那个网站。（解释一下：如果不放在代码块里，LINUX的L就会大写，就打开了错误的网站）2. 安装过程中可能会出现缺少openssl，就用下边这两条语句安装。在ubuntu上安装torch7按照下边的网站来就OK。3. itorch安装参考以下网站的教程。测试一下咯～～然后，在命令行中直接输入。吼开森，感觉自己棒棒哒～]]></description><category></category></item><item><title><![CDATA[use cat to install software list in txt]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/121095202</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/121095202</guid><author>Sun7_She</author><pubDate>Tue, 02 Nov 2021 10:21:21 +0800</pubDate><description><![CDATA[for req in $(cat requirements.txt); do pip install $req; done
]]></description><category></category></item><item><title><![CDATA[如何在linux中创建虚拟环境]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/120846854</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/120846854</guid><author>Sun7_She</author><pubDate>Tue, 19 Oct 2021 15:13:03 +0800</pubDate><description><![CDATA[如何在linux中创建虚拟环境 - 汪凡 - 博客园
]]></description><category></category></item><item><title><![CDATA[ubuntu18.04 cuda8.0 cudnn5 caffe python2.7 install]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/120553127</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/120553127</guid><author>Sun7_She</author><pubDate>Thu, 30 Sep 2021 18:51:07 +0800</pubDate><description><![CDATA[1.

sudo apt install vim

sudo vim /etc/apt/sources.list

ubuntu镜像-ubuntu下载地址-ubuntu安装教程-阿里巴巴开源镜像站

sudo apt-get update

sudp apt-get upgrade

sudo apt install python



2. cuda8.0+cudnn5 install

sudo dpkg -i ...

cuda-license-8-0
cuda-misc-headers-8-0
cu]]></description><category></category></item><item><title><![CDATA[ubuntu make iso 镜像]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/120547244</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/120547244</guid><author>Sun7_She</author><pubDate>Wed, 29 Sep 2021 14:01:38 +0800</pubDate><description><![CDATA[systemback是一个很好的备份还原软件。用它还可以生成自己的linux系统镜像文件。但是在使用时会发现当生成的sblive文件大于4G的时候是没有办法生成iso文件的。这是由于iso文件自身的限制，iso9600对于文件有限制，单个文件不能超过2G，总的iso文件不能超过4G。

但是我们经常见到的系统安装iso文件很多都是超过4G的。这是怎么回事呢。经过我的苦苦搜索终于找到了一点线索。对于大于4G的可以采用udf文件系统。下面就是具体的解决方法了。

解压 .sblive 文件:

mkdir sb]]></description><category></category></item><item><title><![CDATA[ValueError: numpy.ufunc has the wrong size, try recompiling. Expected 192, got 216]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/120507260</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/120507260</guid><author>Sun7_She</author><pubDate>Mon, 27 Sep 2021 14:04:41 +0800</pubDate><description><![CDATA[error

Traceback (most recent call last):
 File "./tools/train_net.py", line 13, in &lt;module&gt;
 from fast_rcnn.train import get_training_roidb, train_net
 File "/home/ruoqi256/srq/model/faster-rcnn-resnet/tools/../lib/fast_rcnn/train.py", line 1...]]></description><category></category></item><item><title><![CDATA[nvidia驱动问题解决 -- Failed to initialize NVML: Driver/library version mismatch]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/73025414</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/73025414</guid><author>Sun7_She</author><pubDate>Mon, 27 Sep 2021 14:02:50 +0800</pubDate><description><![CDATA[dpkg -l | grep -i nvidia]]></description><category></category></item><item><title><![CDATA[pydensecrf使用方法以及各个参数含义]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/86471565</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/86471565</guid><author>Sun7_She</author><pubDate>Mon, 27 Sep 2021 14:02:09 +0800</pubDate><description><![CDATA[pydensecrf使用方法以及各个参数含义

https://github.com/lucasb-eyer/pydensecrf

该模型的一元势能包含了图像的形状、纹理、位置和颜色，二元势能使用了对比度敏感的的双核势能，CRF的二元势函数一般是描述像素点与像素点之间的关系，鼓励相似像素分配相同的标签，而相差较大的像素分配不同标签，而这个“距离”的定义与颜色值和实际相对距离有关，这样CRF能够使...]]></description><category></category></item><item><title><![CDATA[u-net相关论文阅读]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/116710411</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/116710411</guid><author>Sun7_She</author><pubDate>Mon, 27 Sep 2021 14:00:46 +0800</pubDate><description><![CDATA[1. u-net：医疗影像中，器官本身结构固定，语义信息没有特别丰富，所以高级语义信息和低级特征都显得很重要（u-net的skip connection和u形结构就派上了用场）。原始u-net参数量在28m左右，如果把channel数乘被缩小，模型可以更小，缩小两倍后可以是7.75m，缩小4倍，模型参数量可以控制在2m以内，非常轻量
...]]></description><category></category></item><item><title><![CDATA[在线caffe可视化工具]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/119924374</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/119924374</guid><author>Sun7_She</author><pubDate>Thu, 26 Aug 2021 09:36:11 +0800</pubDate><description><![CDATA[http://ethereon.github.io/netscope/#/editor
]]></description><category></category></item><item><title><![CDATA[泛在操作系统]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/116351907</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/116351907</guid><author>Sun7_She</author><pubDate>Sun, 02 May 2021 11:15:06 +0800</pubDate><description><![CDATA[对于泛在操作系统感兴趣的同学们可以看一下下面的博文哦~

https://blog.youkuaiyun.com/AIIT_Ubiquitous

网站源码链接，欢迎贡献IDEA哦~

https://forgeplus.trustie.net/projects/xuos/xuos-web
]]></description><category></category></item><item><title><![CDATA[illegal instruction]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/115297970</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/115297970</guid><author>Sun7_She</author><pubDate>Mon, 29 Mar 2021 12:05:26 +0800</pubDate><description><![CDATA[在Jetson上安装完numpy，pytorch等后，import 模块时出现Illegal instruction错误，后经过pip3 uninstall numpy后错误解决。jetson预装了多个版本的python3，卸载完numpy后还有旧版本的numpy，怀疑numpy重复导致。
————————————————
版权声明：本文为优快云博主「Time-Net」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.youkuaiyun.com/]]></description><category></category></item><item><title><![CDATA[2021-03-02]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/114289166</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/114289166</guid><author>Sun7_She</author><pubDate>Tue, 02 Mar 2021 16:57:52 +0800</pubDate><description><![CDATA[ubuntu配置ip地址详细步骤

https://jingyan.baidu.com/article/e75057f21b6fd7ebc81a8974.html

[Ubuntu]Linux 下支持exFAT（移动硬盘）

https://blog.youkuaiyun.com/weixin_40117614/article/details/83650068
]]></description><category></category></item><item><title><![CDATA[【深度学习】pytorch编写新的层]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/100676005</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/100676005</guid><author>Sun7_She</author><pubDate>Mon, 09 Sep 2019 21:28:35 +0800</pubDate><description><![CDATA[如何在CUDA中为Transformer编写一个PyTorch自定义层

https://cloud.tencent.com/developer/article/1419577

Pytorch拓展进阶(一)：Pytorch结合C以及Cuda语言

https://oldpan.me/archives/pytorch-combine-c-and-cuda






...]]></description><category></category></item><item><title><![CDATA[【深度学习】CVPR2019-PAPER IDEAS]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/93474139</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/93474139</guid><author>Sun7_She</author><pubDate>Mon, 24 Jun 2019 11:47:08 +0800</pubDate><description><![CDATA[1. driving stereo: a large-scale dataset for stereo matching in autonomous driving scenarios

这位朋友的工作主要是在制作深度数据集，制作数据集的过程写了这篇paper。主要是深度标签不好获得，直接用雷达获取的还不够准确，他又结合了其他信息来制作更准确的label～（第一次知道，制作label的方法也可以写...]]></description><category></category></item><item><title><![CDATA[CVPR 2019 新文阅读]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/91892908</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/91892908</guid><author>Sun7_She</author><pubDate>Thu, 13 Jun 2019 23:11:13 +0800</pubDate><description><![CDATA[1 learning a deep convnet for multi-label classification with partial labels

深度学习在单标签分类任务中表现很好，但是日常生活中的图像本质上是多标签的。多标签分类比单标签分类更困难，因为输入图像和输出标签空间都更复杂。与单标签相比，大规模收集干净的多标签注释更难。为了降低标注成本，我们建议训练带有部分标签的模型，即每个图...]]></description><category></category></item><item><title><![CDATA[【深度学习】分类网络结构RESNET RESNEXT DENSENET DPN MOBILE NET SHUFFLE NET]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/90476530</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/90476530</guid><author>Sun7_She</author><pubDate>Fri, 24 May 2019 09:21:23 +0800</pubDate><description><![CDATA[RESNET

跳跃连接



RESNEXT

拆分-转换-合并

增加基数（独立路径的数量）来提高准确度比网络加深或扩大来提高准确度更有效。（为什么呢？）



DENSENET

相加改为相并联

加强shortcut，将所有层直接连接在一起



DPN

将残差通道和densely connected path相融合，实现优缺互补

------------------------...]]></description><category></category></item><item><title><![CDATA[【深度学习】目标检测网络结构SSD RETINANET]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/90409774</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/90409774</guid><author>Sun7_She</author><pubDate>Tue, 21 May 2019 16:37:01 +0800</pubDate><description><![CDATA[one stage

SSD:

SSD是Faster-RCNN和YOLO中做了一次的分类和检测过程放在不同的图像大小上做了多次



RETINANET:

RESNET+FPN+FOCAL LOSS


]]></description><category></category></item><item><title><![CDATA[【深度学习】目标检测网络结构MASK RCNN FPN]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/90407103</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/90407103</guid><author>Sun7_She</author><pubDate>Tue, 21 May 2019 15:55:21 +0800</pubDate><description><![CDATA[其相对于FASTER RCNN改进如下：

1. 分割，检测，分类同时进行

2. 引入ROI ALIGN（对分类影响不大，对分割影响大，因为ROI POOLING对应回图像中的像素会有偏差，该方法会相对准确）代替faster rcnn中的ROI POOLING

ROI ALIGN:https://www.cnblogs.com/wangyong/p/8523814.html

对于检测图片中...]]></description><category></category></item><item><title><![CDATA[【深度学习】目标检测网络结构SPP FAST-RCNN FASTER-RCNN]]></title><link>https://blog.youkuaiyun.com/Sun7_She/article/details/90299702</link><guid>https://blog.youkuaiyun.com/Sun7_She/article/details/90299702</guid><author>Sun7_She</author><pubDate>Sun, 19 May 2019 10:32:07 +0800</pubDate><description><![CDATA[在讲fast-rcnn之前，我们先来看一下spp net吧？～

spp net对r-cnn的改进主要有两点：

    1. 只对原图提取一次特征。输入是图片，在feature层对应位置找到候选框的位置。

    2. 结合空间金字塔方法实现cnns的对不同尺度的输入。实现数据的多尺度输入。分别划分不同的小块对feature进行pooling。

对比r-cnn提速100倍左右...]]></description><category></category></item></channel></rss>