How about if the input is DWF file instead of image PNG file

[DongDem]

Hello @neka-nat san,
I find this repository very interesting. I’d like to ask whether, if I use a DWG file as the input instead of a PNG file, I can keep the current workflow and only change the input-reading step from PNG to DWG before passing it to the LLMs.

[neka-nat]

Since DWG files cannot currently be handled directly, it is necessary to render the DWG using some method and save it as an image for use.
You can use Autodesk Viewer, and I have also developed my own DWG rendering library.
https://github.com/neka-nat/ezdwg

[DongDem]

Hi @neka-nat san,
Sorry for late reply. I want to use file DXF to guarantee the exact information from input instead image data. Now I am converting the file DWG to file DXF using ODAConverter, then use ezdxf to read the file dxf before feed to LLM, but I have the problem, the file dxf contains the content list elements but it is separate so it is very difficult for LLM understand whole block. Do you have any idea for this process?
Thank you!