NOEXIST

> > 各位开发者好，我是 [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) 的作者。 > > Umi-OCR 是一个开源的OCR软件，目前正在开发PDF扫描件识别的功能。其中的一个难点在于，OCR得到的文本块的顺序，往往与实际阅读顺序不符合，特别是在多栏布局的文档中。我需要根据文档的排版，正确区分出不同列，按实际阅读顺序为文本块进行排序。 > > pdf2docx 中也涉及一些基于规则的排版解析功能。我浅读了部分代码，这给了我一些启发。 > > 最终，我设计出一个新算法： [GapTree_Sort 间隙树排序法](https://github.com/hiroi-sora/GapTree_Sort_Algorithm) 。它通过寻找文本块之间的间隙，将页面切割为不同的纵向区块，构建出布局树。最后，前序遍历布局树，即可得到符合人类阅读习惯的文本排序。 > > 当然，除了排序文本块，也能通过布局树分析更多排版信息。（不过它不是针对PDF设计的，没有考虑块对象本身附带的标签等信息。） > > pdf2docx 当前的规则匹配，只支持最多2栏、且列宽不能相差太大。 > > 而 GapTree_Sort...

>  可以先临时异常处理一下 我发现可以用pil_tobytes 但是不确定完不完美 

#298 试试这个pr的代码，手动改改对应库代码

> Sorry for being late, you can simply ignore that messages as they're just warnings in the x64 version, it's working for me tho :/ Warning is another matter,This test...

> Uh I tried with the [latest version](https://github.com/sudo-nautilus/FFmpeg-Builds-Win32/releases/download/latest/ffmpeg-master-latest-win32-gpl.zip), > > 2022-07-18.11-05-40.mp4 I don't know what happened. I tried the new [version.](https://github.com/sudo-nautilus/FFmpeg-Builds-Win32/releases/download/latest/ffmpeg-master-latest-win32-gpl.zip) The virtual machine test I tried to use win11...

> Uhh I too don't have any idea what's happening, anyway how did you manage to install a 32-bit win11 vm I used 64 bit system for testing. Maybe it's...

> > Uhh I too don't have any idea what's happening, anyway how did you manage to install a 32-bit win11 vm > > fwiw, there is no 32-bit Windows...

> > > Uhh I too don't have any idea what's happening, anyway how did you manage to install a 32-bit win11 vm > > > > > > fwiw,...

may be is mtk cpu can't support frame as encode size= phone size try to this: scrcpy -s 0e16808b0501 -V debug -p 21300:22300 -m 1920

测试样本 [test.pdf](https://github.com/user-attachments/files/15926382/test.pdf)