<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[Hi_maxin的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/Hi_maxin</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; Hi_maxin]]></copyright><item><title><![CDATA[libFuzzer教程]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/106391870</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/106391870</guid><author>Hi_maxin</author><pubDate>Wed, 27 May 2020 22:23:00 +0800</pubDate><description><![CDATA[环境配置
$ sudo apt update 
$ sudo apt-get install clang-6.0 
$ update-alternatives --install /usr/bin/clang++ clang++ /usr/bin/clang++-6.0 1000 
$ update-alternatives --install /usr/bin/clang clang /usr/bin/clang-6.0 1000 
$ update-alternatives --config clang]]></description><category></category></item><item><title><![CDATA[linux下shell中使用上下键翻出历史命名时出现^[[A^[[A^[[A^[[B^[[B的问题解决]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/100942859</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/100942859</guid><author>Hi_maxin</author><pubDate>Tue, 17 Sep 2019 22:46:55 +0800</pubDate><description><![CDATA[1：查看当前用户的shell类型命令： echo $SHELL
2：改变当前用户登录默认的shell：sudo chsh -s /bin/bash username

]]></description><category></category></item><item><title><![CDATA[FileZilla尝试连接“ECONNREFUSED - 连接被服务器拒绝”失败]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/100935846</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/100935846</guid><author>Hi_maxin</author><pubDate>Tue, 17 Sep 2019 18:58:11 +0800</pubDate><description><![CDATA[https://blog.youkuaiyun.com/maoyuanming0806/article/details/78539655

]]></description><category></category></item><item><title><![CDATA[Mysql创建用户并授权]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/100565341</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/100565341</guid><author>Hi_maxin</author><pubDate>Thu, 05 Sep 2019 19:27:27 +0800</pubDate><description><![CDATA[create user'mx'@'localhost' identified by '123'
grant all privileges on *.* to 'mx'@'localhost';


]]></description><category></category></item><item><title><![CDATA[Java中List, Integer[], int[]的相互转换]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/97888236</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/97888236</guid><author>Hi_maxin</author><pubDate>Wed, 31 Jul 2019 09:42:20 +0800</pubDate><description><![CDATA[Java中List, Integer[], int[]的相互转换
转自：https://www.cnblogs.com/cat520/p/10299879.html
import java.util.Arrays;
import java.util.List;
import java.util.stream.Collectors;
 
public class Main {
    public ...]]></description><category></category></item><item><title><![CDATA[论文阅读-(CCNet)Criss-Cross Attention for Semantic Segmentation]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/89460746</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/89460746</guid><author>Hi_maxin</author><pubDate>Mon, 22 Apr 2019 20:58:33 +0800</pubDate><description><![CDATA[相关论文 non local 操作


要点:
criss-cross attention module 获得上下文信息
循环操作, 每一个像素可以获得全图信息


优势:
与non-local 操作相比,少了85%的浮点运算,
与non-local 操作相比,GPU占用率时1/11
state of the art


现存问题
Due to the fixed geometric struct...]]></description><category></category></item><item><title><![CDATA[论文阅读- Non-local Neural Networks]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/89460095</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/89460095</guid><author>Hi_maxin</author><pubDate>Mon, 22 Apr 2019 20:27:58 +0800</pubDate><description><![CDATA[现存问题
Both convolutional and recurrent operations are building blocks that process one local neighborhood at a time.(卷积操作只处理neighborhood一小块)


如何捕获远距离的依赖/相关性?

对于序列化的数据,recurrent operations 是主流的解决方案
...]]></description><category></category></item><item><title><![CDATA[压缩文件的分卷压缩和解压]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/88899035</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/88899035</guid><author>Hi_maxin</author><pubDate>Fri, 29 Mar 2019 18:06:14 +0800</pubDate><description><![CDATA[原理上是跟文件夹的分卷压缩是一样的，有些大的文件压缩后还是大，不方便拷贝，这就可以用到分卷压缩，然后再拷贝就方便很多了。我通过实例的方式去讲解实现方法。
例：Ubuntu下有一文件夹myfile，大小约为90G。
直接对文件夹分卷压缩方法：
$ tar cvzpf - myfile | split -d -b 10240m - myfile.
文件夹会以每10G的大小被分卷压缩，生成myfile....]]></description><category></category></item><item><title><![CDATA[Ubuntu Tmux 启用鼠标滚动]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/88716051</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/88716051</guid><author>Hi_maxin</author><pubDate>Thu, 21 Mar 2019 15:15:22 +0800</pubDate><description><![CDATA[按完前缀ctrl+B后，再按冒号：进入命令行模式，
输入以下命令：

set -g mouse on


]]></description><category></category></item><item><title><![CDATA[Pytorch并行训练 示例]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/88245184</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/88245184</guid><author>Hi_maxin</author><pubDate>Wed, 06 Mar 2019 17:31:47 +0800</pubDate><description><![CDATA[pytorch 多卡训练示例
import torch
import torch.nn as nn
from torch.utils.data import DataLoader,Dataset

# parameter and DataLoaders
input_size = 5
output_size = 2
batch_size = 30
data_size = 100

device = ...]]></description><category></category></item><item><title><![CDATA[deeplab中 global average pooling(GAP) 原理]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/88243736</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/88243736</guid><author>Hi_maxin</author><pubDate>Wed, 06 Mar 2019 17:08:31 +0800</pubDate><description><![CDATA[之前有疑惑deeplab中ASPP(Atrous Spatial Pyramid Pooling) 采用多尺度空洞卷积＋GAP并联提取多尺度语义信息时,GAP得到的结果是一维的,而空洞卷积得到的结果时二维的,不知道其中的细节是怎么操作从而实现并联的
查资料＋看源码
资料
https://sthalles.github.io/deep_segmentation_network/写到
为了增加全局信息...]]></description><category></category></item><item><title><![CDATA[修改gcc/g++优先级   多版本gcc/g++ 共存]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/88055277</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/88055277</guid><author>Hi_maxin</author><pubDate>Fri, 01 Mar 2019 13:10:36 +0800</pubDate><description><![CDATA[set priority to use [g++/gcc-4.9] before [g++/gcc-5] .
–&amp;amp;gt;安装 gcc-4.9 and g+±4.9
~$ sudo apt-get install gcc-4.9 g+±4.9
–&amp;amp;gt; 设置优先级: “priority=100” for gcc-4.9 and “priority=50” for gcc-5.
~$ sudo up...]]></description><category></category></item><item><title><![CDATA[图论分割与全景分割]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/86471735</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/86471735</guid><author>Hi_maxin</author><pubDate>Mon, 14 Jan 2019 10:57:39 +0800</pubDate><description><![CDATA[图论分割与全景分割

传统图分割算法相关工作
预计在2.10（春节假期结束为止）完成相关工作，尽可能提前完成工作
工作目标：（春节假期结束前完成 （20天左右））


明确问题的定义


明确图分割算法的评价标准（1天）



理解总结各个算法的核心思想 （15天）
总结算法中存在的问题以及思考改进措施 （与上一步同步进行）
挑选比较好的算法（多个）进行复现   （5天）

全景分割相关工作
工作...]]></description><category></category></item><item><title><![CDATA[pyinstaller打包问题]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/86363364</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/86363364</guid><author>Hi_maxin</author><pubDate>Sat, 12 Jan 2019 15:58:43 +0800</pubDate><description><![CDATA[pyinstaller -F -w main.py
-F是打成一个exe
-W是不弹黑框
问题1

PyInstaller\compat.py&quot;, line 370, in
exec_command
out = out.decode(encoding)
UnicodeDecodeError: ‘utf-8’ codec can’t decode byte 0xff in position 152:...]]></description><category></category></item><item><title><![CDATA[Google Cloud SDK配置 解决ERROR: Reachability Check failed问题]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/85697348</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/85697348</guid><author>Hi_maxin</author><pubDate>Thu, 03 Jan 2019 15:08:53 +0800</pubDate><description><![CDATA[安装google cloud sdk的时候第一次需要初始化，而初始化的时候因为一些原因，
会报错:
Network diagnostic detects and fixes local network connection issues.
Checking network connection...done.                                             ...]]></description><category></category></item><item><title><![CDATA[Faster R-CNN系列（四）：RPN的实现之神经网络部分的实现]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/85322386</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/85322386</guid><author>Hi_maxin</author><pubDate>Fri, 28 Dec 2018 15:08:38 +0800</pubDate><description><![CDATA[RPN的实现之神经网络部分的实现
RPN的框架：



图解

上支路是所有anchor的修正量（4*9）
下支路是前景背景9个anchor的分类得分(前景 背景)



rpn_match

label为1的anchor： 当一个anchor与真实bounding box的最大IOU超过阈值Vt1(0.7)
label为-1的anchor ： 当一个anchor与真实bounding box的最...]]></description><category></category></item><item><title><![CDATA[实现自己的Faster RCNN（Keras） （step-by-step）]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/85320417</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/85320417</guid><author>Hi_maxin</author><pubDate>Fri, 28 Dec 2018 13:12:24 +0800</pubDate><description><![CDATA[Faster R-CNN系列的汇总：
运行开源版

1.window 运行开源的tensorflow版 faster rcnn

自己实现keras版faster rcnn （step-by-step）

以下两篇为keras 与faster rcnn的基础理解与实验

2.anchor 的实现
3.用keras构建多输入-多输出-多自定义loss的模型


更新中


...]]></description><category></category></item><item><title><![CDATA[Faster R-CNN系列（三）：用keras构建多输入-多输出-多自定义loss的模型]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/85281415</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/85281415</guid><author>Hi_maxin</author><pubDate>Thu, 27 Dec 2018 14:33:32 +0800</pubDate><description><![CDATA[用keras构建多输入-多输出-多自定义loss的模型
keras 有两种模型

Sequential 顺序模型
Model 函数式模型  Model(inputs,outputs,name=None)

from keras import Sequential, Model
import keras.layers as KL
import keras.backend as K
import nu...]]></description><category></category></item><item><title><![CDATA[LeetCode 个人解答记录(11-15)]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/85275451</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/85275451</guid><author>Hi_maxin</author><pubDate>Thu, 27 Dec 2018 11:46:52 +0800</pubDate><description><![CDATA[LeetCode 个人解答记录(11-15)


Container With Most Water
双指针，从两端开始，移动较短的那个边向中间靠拢，寻找过程中的最优解

Runtime: 16 ms, faster than 60.82% of C++ online submissions for Container With Most Water.

   class Solution {
 ...]]></description><category></category></item><item><title><![CDATA[Faster R-CNN系列（二）：实现anchor]]></title><link>https://blog.youkuaiyun.com/Hi_maxin/article/details/85248141</link><guid>https://blog.youkuaiyun.com/Hi_maxin/article/details/85248141</guid><author>Hi_maxin</author><pubDate>Tue, 25 Dec 2018 15:02:59 +0800</pubDate><description><![CDATA[Faster R-CNN系列（二）：实现anchor

导入相关的包

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.patches as patches
%matplotlib inline


定义feature map 的size
在这里定义为16*16
每个feature map上的点对应原图中8*...]]></description><category></category></item></channel></rss>