Workaround about diagram recognizing

[dantetemplar]

I encountered a problem processing diagrams while reading presentations at my university.

Can you tell me where in the code I could intercept the diagram processing? I would like to add a fallback to a computer vision model so that the model's response is inserted instead of the diagram block.

By diagram I mean something like this:

So far I got complete mess:

#### Half Subtractor and Half Adder comparison

𝑎
!
𝑏 !

𝑎
!
𝑏 !

𝑑
!

𝑠
!

XOR

XOR

NOT
𝑏out

AND

NOT
𝑏out

Half Subtractor Circuit:
-  No borrow-in from outside

-  The Subtraction of Least Significant Bit

𝑐out
AND

Half Adder Circuit:
-  No carry-in from outside

-  The Summation of Least Significant Bit

-----

[JorjMcKie]

Can you let me have an example page please? Or try parameter "write_images=True". This option should spare out vector graphics and images from text extraction, write the respective areas out to image files (PNG) and insert Markdown references to these images in the produced text.

[dantetemplar]

Can you let me have an example page please? Or try parameter "write_images=True". This option should spare out vector graphics and images from text extraction, write the respective areas out to image files (PNG) and insert Markdown references to these images in the produced text.

Now, I do it that way: find all areas with images and non-table graphics clusters, skip text processing in these areas; make "screenshot" (create pixmap) for these areas and pass it to tesseract ocr. And also insert Markdown references.

[dantetemplar]

We can close the issue actually. I will send my code snippets later

[dantetemplar]

https://github.com/dantetemplar/pymupdf4llm