Package versions

[francesco-cardinale]

Hi everyone,

I'm trying to set up the project, but I'm running into issues due to the lack of specified versions in requirements.txt. Without version constraints, dependencies might not be compatible, leading to unexpected errors.

For example, when attempting to preprocess an image with text, I get this error:
ImportError: cannot import name '_imaging' from 'PIL'

I suspect this is related to an incompatible version of Pillow or another dependency.

My Questions:

1. Has anyone successfully set up the project recently? If so, which package versions are you using?
2. Could the maintainers provide a list of tested versions or a pip freeze output?
3. Would it make sense to update requirements.txt with specific versions to avoid compatibility issues?

Thanks in advance for any help!

[FWao]

Hi,

thank you for pointing this out, I will include a versioned requirements file in the future when I have time.

For now I can give you a freeze of a fresh environment which I just created on MacOS using the requirements.txt file (Python 3.12.9):
requirements.freeze.txt

Maybe your Python or Pillow version is too old?

[francesco-cardinale]

Hi @FWao,

I'm using Python 3.12.8 and the same package versions as in the provided requirements.freeze.txt, but I'm still encountering the same issue when preprocessing a PNG file containing text (and also with other file types).

I'm running the project using "Run with Docker (API only)" since my machine isn't very powerful.

Here’s the error log:

web-1  | Running in API-only mode. No llama-cpp server will be loaded.
web-1  | Initializing OpenAI client with API URL:  https://api.openai.com/v1
web-1  | Start Webserver on http://0.0.0.0:5000                                                                                        
web-1  | Please use http://localhost:5000 to access the web app locally or the IP / hostname of your server to access the web app in your local network.                                                                                                                      
web-1  |  * Debug mode: on
web-1  | Traceback (most recent call last):
web-1  |   File "/usr/bin/ocrmypdf", line 5, in <module>
web-1  |     from ocrmypdf.__main__ import run
web-1  |   File "/usr/lib/python3/dist-packages/ocrmypdf/__init__.py", line 10, in <module>
web-1  |     from ocrmypdf import helpers, hocrtransform, pdfa, pdfinfo
web-1  |   File "/usr/lib/python3/dist-packages/ocrmypdf/helpers.py", line 20, in <module>
web-1  |     import img2pdf
web-1  |   File "/usr/lib/python3/dist-packages/img2pdf.py", line 25, in <module>
web-1  |     from PIL import Image, TiffImagePlugin, GifImagePlugin
web-1  |   File "/app/.venv/lib/python3.12/site-packages/PIL/Image.py", line 100, in <module>
web-1  |     from . import _imaging as core
web-1  | ImportError: cannot import name '_imaging' from 'PIL' (/app/.venv/lib/python3.12/site-packages/PIL/__init__.py)
web-1  | Traceback (most recent call last):
web-1  |   File "/app/webapp/input_processing/routes.py", line 401, in preprocess_file
web-1  |     with fitz.open(ocr_output_path) as ocr_pdf:
web-1  |          ^^^^^^^^^^^^^^^^^^^^^^^^^^
web-1  |   File "/app/.venv/lib/python3.12/site-packages/pymupdf/__init__.py", line 2955, in __init__
web-1  |     raise FileNotFoundError(msg)
web-1  | pymupdf.FileNotFoundError: no such file: '/tmp/tmpqvt9lckv/ocr_OCR-test-2.pdf'

Thanks in advance for any insights!

[FWao]

Thanks for pointing out the malfunctioning docker container. I see this behavior too and don't think uploading images ever worked with the api-only docker container. I guess this is because of some dependency issues inside of the container which I still have to figure out as it works without a problem without using Docker.

[FWao]

@francesco-cardinale Can you test it with the current api-only docker image? Right now I can just test arm64 image, but it seems to work now.

[francesco-cardinale]

@FWao I get the same error if I try, for example, to upload the attached PDF file in 'Document Preprocessing'
ocr-test-pdf.pdf

[FWao]

On which OS/platform are you running the docker image?

[francesco-cardinale]

Windows 11 Pro

[FWao]

Do you use tesseract as the OCR method?

I tried the current api-only docker image again on a linux machine and they work there too with your provided file using tesseract. The behavior on Windows should be the same as docker is running in wsl2.

Please make sure you are really running the current version of the docker image:

This should pull a new package as I rebuilt them today: docker compose -f docker-compose-api.yml pull