setup instructions for Windows

[freecho]

Please check the relevant content in the Windows section of the readme.md.

[VikParuchuri]

Thanks for this! This doesn't seem to include some dependencies:

• ocrmypdf
• tesseract + language packs
• ghostscript

Did you find how to install these?

[freecho]

Thanks,I will continue to study these issues when free time.

[GdeJoode]

Hi, I found the tesseract 5 binary on https://github.com/UB-Mannheim/tesseract/wiki. I am just now installing marker and am waiting for a nvidia graphics card to use CUDA. So have not tested it yet.

or winget install -e --id UB-Mannheim.TesseractOCR (found through the OCRmyPDF page)

[CBIhalsen]

Reference in new issue

HI, have you found the perfect solution for windows?

[freecho]

I have updated the README.md, and following the steps, both my friend and I have successfully run the project.

[umarbutler]

I encountered two issues following the steps detailed in your instructions:

1. I encountered the error ImportError: failed to find libmagic. Check your installation which I resolved by running (in poetry shell) pip install python-magic-bin==0.4.14 as described in this StackOverflow answer.
2. I encountered the error AssertionError: Torch not compiled with CUDA enabled which I resolved by running (in poetry shell) pip uninstall torch and then pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118 (note that this will install torch complied for CUDA 11.8) as described by PyTorch's official documentation.
3. ModuleNotFoundError: No module named 'pkg_resources' which I resolved by running pip install setuptools.

After running these fixes, I was able to get marker working in my Windows environment.

[umarbutler]

I have created a new PR that addresses the issues raised in my comment above.

[VikParuchuri]

Thanks for looking into this! The new version has no required system dependencies, so Windows install should be much simpler - https://github.com/VikParuchuri/marker/pull/116