<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[Treasure Ashes]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/Treasureashes</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; Treasureashes]]></copyright><item><title><![CDATA[如果你使用了＜iframe＞但是不显示内容]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/130234922</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/130234922</guid><author>Treasureashes</author><pubDate>Wed, 19 Apr 2023 03:01:45 +0800</pubDate><description><![CDATA[iframe内嵌的网页链接复制到浏览器可以看到内容，但是用iframe看不到。你的链接在前面没有加http://。localhost也要加。]]></description><category></category></item><item><title><![CDATA[DataBricks jar包写的相对路径文件找不到]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/130234912</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/130234912</guid><author>Treasureashes</author><pubDate>Wed, 19 Apr 2023 02:56:58 +0800</pubDate><description><![CDATA[因为它源代码的逻辑是在运行目录创建一个results文件夹，再把所有的以expectedName命名的results文件写到里面（folderPath+expName+suffix），但是谁知道DataBricks运行jobs的目录在DBFS哪里。写到这里我突然想到或许可以尝试每次新建新的job来上传新的jar包，但懒得再尝试了，这个结论也是我尝试了多次打包上传才发现的。anyway，我突然灵机一动，既然jar包内部是一个black box我动不了，那我可以在参数上做文章啊。我可真是一个大天才！]]></description><category></category></item><item><title><![CDATA[Golang 如何不用uuid的包产生一个random的uuid]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/129919680</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/129919680</guid><author>Treasureashes</author><pubDate>Sun, 02 Apr 2023 23:45:21 +0800</pubDate><description><![CDATA[【代码】Golang 如何不用uuid的包产生一个random的uuid。]]></description><category></category></item><item><title><![CDATA[Golang 如何得到git repo 的 URL]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/129919765</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/129919765</guid><author>Treasureashes</author><pubDate>Sun, 02 Apr 2023 23:39:45 +0800</pubDate><description><![CDATA[【代码】Golang 如何得到git repo 的 URL。]]></description><category></category></item><item><title><![CDATA[Golang 如何判断一个podman container是否在running]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/129919727</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/129919727</guid><author>Treasureashes</author><pubDate>Sun, 02 Apr 2023 23:37:24 +0800</pubDate><description><![CDATA[【代码】Golang 如何判断一个podman container是否在running。]]></description><category></category></item><item><title><![CDATA[Golang os.Write 不是 atomic 的]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/129919567</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/129919567</guid><author>Treasureashes</author><pubDate>Sun, 02 Apr 2023 23:29:28 +0800</pubDate><description><![CDATA[Golang os.Write 不是 atomic 的，在parallel process 中容易产生 write read race。更好的做法是先写到一个temp file里，再os.Rename()]]></description><category></category></item><item><title><![CDATA[Script Identification of Multi-Script Documents: A Survey]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/125840153</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/125840153</guid><author>Treasureashes</author><pubDate>Mon, 18 Jul 2022 03:19:42 +0800</pubDate><description><![CDATA[这是一篇2017年来自新疆大学的Survey，写得不是很好，没有提出什么观点，没有很好将各种方法串连起来，各种分类阐述也很粗暴。而且由于是2017年的文章，其中对当前研究的工作的总结已经不太具有太多的价值。但是其它方面的整理分类工作还是有一定价值，适合对script identification（脚本识别）这个领域没有太多了解的人了解一下。...]]></description><category></category></item><item><title><![CDATA[Text Recognition in the Wild: A Survey]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/125840148</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/125840148</guid><author>Treasureashes</author><pubDate>Mon, 18 Jul 2022 03:12:03 +0800</pubDate><description><![CDATA[这是来自华南理工2020年的一篇survey，主要讨论的是场景文本识别（scene text recognition，STR）领域的现状与一些发展方向。]]></description><category></category></item><item><title><![CDATA[Hamming OCR: A Locality Sensitive Hashing Neural Network for Scene Text Recognition]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/125840145</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/125840145</guid><author>Treasureashes</author><pubDate>Mon, 18 Jul 2022 03:08:56 +0800</pubDate><description><![CDATA[当前的OCR模型太大，直接应用在中文等多字符语言系统的识别中，会导致过大的存储成本（主要是projection matrix的存储）和较大的计算量，如果直接缩减d（字符级特征向量的维度）会损害模型性能，直接缩减L（词汇表的长度）会限制应用场景。...]]></description><category></category></item><item><title><![CDATA[【量化】A Quantization-Friendly Separable Convolution for MobileNets]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/125840142</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/125840142</guid><author>Treasureashes</author><pubDate>Mon, 18 Jul 2022 03:03:32 +0800</pubDate><description><![CDATA[将深度卷积+BN+ReLU+点卷积+BN+ReLU替换成深度卷积+点卷积+BN+ReLU（+L2正则化）]]></description><category></category></item><item><title><![CDATA[【剪枝】ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/125840131</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/125840131</guid><author>Treasureashes</author><pubDate>Mon, 18 Jul 2022 02:58:38 +0800</pubDate><description><![CDATA[剪枝的思想可以被总结为一句话：（以某一标准）评估每一个神经元的重要性，移除不重要的那些神经元，再finetune整个网络来恢复精度。所以，剪枝一般有三个阶段：filter selection，pruning，与fine-tuning。...]]></description><category></category></item><item><title><![CDATA[【量化】Post-Training Quantization for Vision Transformer]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/125840116</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/125840116</guid><author>Treasureashes</author><pubDate>Mon, 18 Jul 2022 02:48:55 +0800</pubDate><description><![CDATA[Ranking aware+Bias correction+Nuclear Norm Based Mixed-Precision Quantization]]></description><category></category></item><item><title><![CDATA[【Backbone】MOBILEVIT: LIGHT-WEIGHT, GENERAL-PURPOSE,AND MOBILE-FRIENDLY VISION TRANSFORMER]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/125840100</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/125840100</guid><author>Treasureashes</author><pubDate>Mon, 18 Jul 2022 02:37:29 +0800</pubDate><description><![CDATA[文章的主要思想是用transformer来代替convolution去进行global processing，而convolution主要是来进行local processing。]]></description><category></category></item><item><title><![CDATA[【量化】PTQ4ViT: Post-Training Quantization Framework for Vision Transformers]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/125840071</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/125840071</guid><author>Treasureashes</author><pubDate>Mon, 18 Jul 2022 02:25:26 +0800</pubDate><description><![CDATA[作者的idea来源于对ViT中激活值的分布的观察，以及对不同评估量化误差的指标对比的观察。]]></description><category></category></item><item><title><![CDATA[【Shell】ln: Too many levels of symbolic links]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/125840062</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/125840062</guid><author>Treasureashes</author><pubDate>Mon, 18 Jul 2022 02:09:30 +0800</pubDate><description><![CDATA[答案在，这里出现了路径的循环引用]]></description><category></category></item><item><title><![CDATA[利用torch.fx进行后量化]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/121048421</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/121048421</guid><author>Treasureashes</author><pubDate>Sat, 30 Oct 2021 11:19:52 +0800</pubDate><description><![CDATA[torch.fx 量化支持——FX GRAPH MODE QUANTIZATION
torch.fx目前支持的量化方式：

Post Training Quantization

Weight Only Quantization
Dynamic Quantization
Static Quantization


Quantization Aware Training

Static Quantization



其中，Post Training Quantization中的Static Quantiza]]></description><category></category></item><item><title><![CDATA[2021-10-30 用torch.fx后量化的bug记录]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/121048372</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/121048372</guid><author>Treasureashes</author><pubDate>Sat, 30 Oct 2021 11:18:33 +0800</pubDate><description><![CDATA[报错：ValueError: input model must be a GraphModule, please run torch._fx.symbolic_trace on your model before using quantize_fx. Got type:&lt;class ‘main.EVAL’&gt;

原因：模型中有不是graph module的部分
解决方案：提供的torch._fx.symbolic_trace来看是哪个部分出问题了：
st = symbolic_trace(net.]]></description><category></category></item><item><title><![CDATA[【Leetcode】动态规划题思路总结（Medium）]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/119723322</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/119723322</guid><author>Treasureashes</author><pubDate>Mon, 16 Aug 2021 00:40:04 +0800</pubDate><description><![CDATA[根据难度分为了三篇，Medium中回文串相关的会单独介绍一篇。

目录

Medium

62. 不同路径

64. 最小路径和

198. 打家劫舍

213.打家劫舍 II

740. 删除并获得点数

1143. 最长公共子序列

最长公共子串

300. 最长递增子序列

375. 摆动序列

Medium

62. 不同路径

爬楼梯的二维版本，不同的是路只能一格一格走

状态：dp[i][j]到达当前格子[i][j]的可能的路径数

递推公式：dp[i][j] = dp[i-1][j] ...]]></description><category></category></item><item><title><![CDATA[【leetcode】动态规划题思路总结（Easy）]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/119722221</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/119722221</guid><author>Treasureashes</author><pubDate>Sun, 15 Aug 2021 23:15:16 +0800</pubDate><description><![CDATA[全部写下来发现太长了，就根据难度分为三篇了。

Easy

70. 爬楼梯 / 509. 斐波那契数

这道题可以算是最经典的动态规划题了

状态：dp[i] 到达第i台阶可能的跳法

递推公式：dp[i] = dp[i-1] + dp[i-2]

初始状态：dp[0] = 0，dp[1] = 1，dp[2] = 2（这个边界条件也可以设为dp[0] = 1，dp[1] = 1）


class Solution(object):
    def climbStairs(self, n):
       .]]></description><category></category></item><item><title><![CDATA[多任务学习优化总结 Multi-task learning（附代码）]]></title><link>https://blog.youkuaiyun.com/Treasureashes/article/details/119003456</link><guid>https://blog.youkuaiyun.com/Treasureashes/article/details/119003456</guid><author>Treasureashes</author><pubDate>Thu, 29 Jul 2021 03:08:52 +0800</pubDate><description><![CDATA[Multi-task Learning(Review)多任务学习概述

多任务学习优化（Optimization in Multi-task learning）



Multi-Task Learning as Multi-Objective Optimization 阅读笔记

Multi-Task Learning Using Uncertainty to Weigh Losses

GradNorm：Gradient Normalization for Adaptive Loss Balancing]]></description><category></category></item></channel></rss>