Google introduced the content ID system in 2007 to cope with the unauthorized use of copyrighted music and video content. The content ID system has successfully helped the copyright owners to find the pirated content and paid billions of dollars to the copyright owners¹.

基于内容的视频检索（或视频指纹），是根据视频内容，从数据库中找到和当前视频在画面上重复高的视频。

无论是推荐还是检索任务，都离不开向量检索引擎，当向量数量巨大的时候，使用精确搜索的时间成本对于用户来说无法接受，必然会用到近似最邻近搜索（approximate nearest neighbor, 简称 ANN)，本文总结了一些常见的实现了 ANN 算法的向量搜索引擎/工具。

I saw on YouTube a video made by TJ DeVries on how to contribute to the neovim project. There is a scene where he commit via git commit, and whoa, a Neovim instance appeared with three split, with top left the commit message area, top right the git-status area, and on the right the git-diff info, which is really cool.

We have a Python web service where we store some key-val pairs in redis. Occasionally, I want to delete some of the keys matching a certain pattern. Current, we are using redis-py-cluster for redis-related operations.

Over the past two years, I have been using vim-plug as my Nvim plugin manager. Vim-plug is powerful, fast and amazing, especially considering that all its features are contained in a single vim script. It uses the old way to manage plugins, i.e., it will manage the loading of plugins directly.

快排是一种快速排序算法，原理是从数组中选择一个 pivot，数组中小于等于 pivot 的元素放到左边，大于 pivot 的元素放到右边。然后对左边和右边数组分别递归进行快速排序，最后整个数组就成为排序好的数组。

I use Python in my work, and rarely need C++. So I need to compile this C++ cheatsheet for common STL containers/adapters just in case I need them.

In C++, we can build a library either statically or dynamically. In this post, I want to summarize how to build a static and a shared/dynamic library¹.

Although std::cout in C++ is powerful, it lacks the ease of use of format string in Python. Fortunately, the 3rd party package fmt provides a similar feature for C++.

Clangd is a language server for C++/C etc. On CentOS 7, I try to use clangd 12 for code auto-completion. Clangd can be easily installed via its binary release.

When I want to build glibc using the provided configure script (./configure --prefix=xxx), I saw the following error message:

In this post, I would like to share how to squash commits using git-rebase.

在 之前的文章 中，我介绍了几款试用的安卓外贸盒子，由于做工以及试用中发现的问题，最终都没有入手。其实品牌电视盒子，也有一个不错的选择，就是当贝 B3 Pro，趁着京东 6.18 活动，最终以 699 价格入手了这款盒子。

This post continues my previous post on nifty Nvim/Vim techniques.

家里有一台闲置的电视，想平时用来看 YouTube，看电影，于是决定入手一款电视盒子。试用了几款价格比较低的品牌电视盒子（小米、当贝），发现都无法轻松观看 YouTube，安装安卓应用也经常装不了。在寻找合适的电视盒子过程中了解到外贸盒子这种品类，外贸盒子的特点是配置高，但是通常做工一般。最近一个月也试用了两款电视盒子，实际上手体验的确不佳。

We have set up a server to convert pptx files to pdf files using LibreOffice (version: 6.0.7.3 ). Libreoffice is started using subprocess.run() command in Python. The external command I use is something like the following¹:

A brief summary of how to read and write excel files using Pandas package.

I have several PDF files that I want to merge to one file. Each page in the PDF is actually an image. Here is what works best for me.