<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[abc20002929的专栏]]></title><description><![CDATA[经世致用]]></description><link>https://blog.youkuaiyun.com/abc20002929</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; abc20002929]]></copyright><item><title><![CDATA[【deeplearning tips】dice loss解决分割正负样本不均衡]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/131270753</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/131270753</guid><author>abc20002929</author><pubDate>Sun, 18 Jun 2023 13:54:56 +0800</pubDate><description><![CDATA[dice loss可以理解为检测任务重的IOU，用于分割任务中可以有效缓解正样本占比小的case。比如边缘的分割、小目标的分割。]]></description><category></category></item><item><title><![CDATA[opencv轮廓提取指定数量角点]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/131078613</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/131078613</guid><author>abc20002929</author><pubDate>Wed, 07 Jun 2023 00:13:03 +0800</pubDate><description><![CDATA[opencv中的approxPolyDP可以用来找轮廓的角点，但是eps参数设置不是很灵活。通过返回距离信息，可灵活提取指定数量角点。]]></description><category></category></item><item><title><![CDATA[pytoch转tensorfow模型pad计算对齐(mmdetection)]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/130644444</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/130644444</guid><author>abc20002929</author><pubDate>Sat, 13 May 2023 15:00:09 +0800</pubDate><description><![CDATA[mmdetection训练 -> mmdeploy转onnx -> onnx2tf转成tflite各部分安装流程，参考各自官方文档即可。]]></description><category></category></item><item><title><![CDATA[vibe背景建模]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/128659406</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/128659406</guid><author>abc20002929</author><pubDate>Thu, 12 Jan 2023 14:40:24 +0800</pubDate><description><![CDATA[vibe背景建模算法]]></description><category></category></item><item><title><![CDATA[高斯、拉普拉斯金字塔C实现]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/127509004</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/127509004</guid><author>abc20002929</author><pubDate>Tue, 25 Oct 2022 10:48:00 +0800</pubDate><description><![CDATA[拉普拉斯金字塔c实现]]></description><category></category></item><item><title><![CDATA[拉普拉斯金字塔分解算法]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/127090217</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/127090217</guid><author>abc20002929</author><pubDate>Wed, 28 Sep 2022 15:24:58 +0800</pubDate><description><![CDATA[拉普拉斯金字塔分解]]></description><category></category></item><item><title><![CDATA[轮廓提取算法]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/78616745</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/78616745</guid><author>abc20002929</author><pubDate>Sun, 08 May 2022 10:40:27 +0800</pubDate><description><![CDATA[cv]]></description><category></category></item><item><title><![CDATA[python解析tflite模型文件]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/123855174</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/123855174</guid><author>abc20002929</author><pubDate>Wed, 30 Mar 2022 20:15:07 +0800</pubDate><description><![CDATA[目的

解析flatbuffer格式的tflite文件，转成可读的python dict格式

方法

   


#/tensorflow/lite/tools/visualize.py
import re
from tensorflow.lite.python import schema_py_generated as schema_fb
 
def BuiltinCodeToName(code):
    """Converts a builtin op code enum to a rea...]]></description><category></category></item><item><title><![CDATA[AE(associative embedding) loss TF版本实现]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/120492086</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/120492086</guid><author>abc20002929</author><pubDate>Sun, 26 Sep 2021 17:03:21 +0800</pubDate><description><![CDATA[一、背景

    AE loss最早出自《Associative Embedding:End-to-End Learning for Joint Detection and Grouping》这篇文章，通过简单的embedding方式，用于bottom-up关键点检测中进行最后的聚类。

在CornerNet、HigherHRNet等文章中的均使用了该技巧。在目标检测任务ConrnerNet中用来区分多个同类别目标，在姿态估计任务HigherHRNet中用来完成关键点聚类。

 ...]]></description><category></category></item><item><title><![CDATA[tflite模型调试 - 中间层output输出]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/112529203</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/112529203</guid><author>abc20002929</author><pubDate>Tue, 12 Jan 2021 15:38:22 +0800</pubDate><description><![CDATA[背景

由于tflite模型设计时考虑到了内存占用等问题，其forward的过程会反复利用中间内存空间，而导致中间层的output数据没有保存下来，模型debug难点较大。

经过查阅相关材料，发现一个最简洁的方式，记录于此。

思路：解析flatbuffer格式的tflite模型， 直接修改tflite模型内容，调整模型output到指定层。

实现

1. 修改tflite模型output的函数。


from tensorflow.lite.python import schema_py_genera]]></description><category></category></item><item><title><![CDATA[shell cmd]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/85318278</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/85318278</guid><author>abc20002929</author><pubDate>Fri, 28 Dec 2018 10:39:28 +0800</pubDate><description><![CDATA[字符串分割

cut -f1 -d'/'

字符串分割2

| awk '{print $2}'

分割后文件名，批量mv

cut -f1 -d'/' | xargs -I {} mv {} ./xxxx/{}

grep文件差异比较(相对粗糙)

cat file1 | grep -v -f file2

 

 

liunx遍历文件


#include &amp;lt;dirent.h&amp;gt;
...]]></description><category></category></item><item><title><![CDATA[【opencv源码剖析】图像分割 floodFill与graphsegmentation]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/79254355</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/79254355</guid><author>abc20002929</author><pubDate>Wed, 07 Feb 2018 18:22:25 +0800</pubDate><description><![CDATA[一、前言

最近在使用floodFill这个算法时，突然想起selective search里的基础图像分割算法 - 基于图的graphsementation。


两者是比较简单的算法，存在相似之处，即都使用了相邻像素之间像素强度关系。
同时又存在不同点，floodFill关注点是像素层面上，生成一个区域；
而graphsementation由像素逐步构建出图块，生成多个区域。
]]></description><category></category></item><item><title><![CDATA[seetaface 人脸检测算法简析]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/75193876</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/75193876</guid><author>abc20002929</author><pubDate>Sat, 15 Jul 2017 22:40:57 +0800</pubDate><description><![CDATA[一、前言
    使用opencv的lbp人脸检测算法准确率、误检率表现均不是很好，另外算法需跑在ARM上，虽然深度学习一些算法如mtcnn效果很好，但速度很难保证。最终，查阅了seetaface的人脸检测算法论文及源码，并做记录总结于此。]]></description><category></category></item><item><title><![CDATA[opencv lbp人脸检测算法]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/70598314</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/70598314</guid><author>abc20002929</author><pubDate>Mon, 24 Apr 2017 23:29:52 +0800</pubDate><description><![CDATA[一、 算法
算法使用的特征为MB-LBP特征



二、 xml文件格式分析]]></description><category></category></item><item><title><![CDATA[【opencv源码剖析】霍夫圆hough circle]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/55549702</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/55549702</guid><author>abc20002929</author><pubDate>Sat, 18 Feb 2017 00:52:58 +0800</pubDate><description><![CDATA[算法的整体思路：
1.根据设定的阈值canny_threshold，使用canny边缘检测得到可能为圆边缘的点edges。
2.分别计算x,y方向的sobel梯度，用来判断edges点的边缘梯度方向，一个点需计算正、反两个梯度方向(由源码中的k1控制)。
3.遍历所有的edges点，根据设定的圆半径范围[min_radius,max_radius]，在累加投票图中统计圆心出现的次数(参数dp]]></description><category></category></item><item><title><![CDATA[opencv eigenfaces]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/54958257</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/54958257</guid><author>abc20002929</author><pubDate>Thu, 09 Feb 2017 21:51:48 +0800</pubDate><description><![CDATA[Eigenfaces

The problem with the image representation we are given is its high dimensionality. Two-dimensional  grayscale
 images span a -dimensional vector space, so an image with  pixels
 lies i]]></description><category></category></item><item><title><![CDATA[opencv实现简单的局部极大值求解]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/47361523</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/47361523</guid><author>abc20002929</author><pubDate>Sat, 08 Aug 2015 20:48:49 +0800</pubDate><description><![CDATA[前言
实现类似matlab的findpeaks功能。
http://cn.mathworks.com/help/signal/ref/findpeaks.html#buhdm28-1


局部极大值求解
//src: 输入数据
//minPeakDistance: 相邻两个局部极大值的最小间距。
//minHeightDiff: kernelSize范围内极大值与极小值的最小差值。与m]]></description><category></category></item><item><title><![CDATA[Machine Learning with OpenCV]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/47173009</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/47173009</guid><author>abc20002929</author><pubDate>Fri, 31 Jul 2015 17:18:00 +0800</pubDate><description><![CDATA[Machine Learning with OpenCV


By Philipp
 Wagner | May 25, 2010

Machine Learning is a branch of Artificial Intelligence and concerned with the question how to make machines able to learn from ]]></description><category></category></item><item><title><![CDATA[【opencv源码剖析】背景建模mog2]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/43247425</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/43247425</guid><author>abc20002929</author><pubDate>Thu, 29 Jan 2015 23:55:00 +0800</pubDate><description><![CDATA[...]]></description><category></category></item><item><title><![CDATA[opencv任意形状ROI]]></title><link>https://blog.youkuaiyun.com/abc20002929/article/details/41016639</link><guid>https://blog.youkuaiyun.com/abc20002929/article/details/41016639</guid><author>abc20002929</author><pubDate>Tue, 11 Nov 2014 20:10:28 +0800</pubDate><description><![CDATA[相关代码，比较简单，一看就懂：]]></description><category></category></item></channel></rss>