lazy import PyMuPDF

[jzhang533]

To address the compatibility issues caused by PyMuPDF dependency.

• license compatibility: LICENSE issue #11634, #Add pip option to skip PyMuPDF has very restrictive license #8873
• env compatibility: 有计划升级PyMuPDF以便支持python3.11吗？ #11119,安装paddleocr时出现错误：PyMuPDF无法安装 #7934, Update requirements.txt #11662
• and you can see more discussions by searing PyMuPDF in issues.

This pull request proposes a lazy import for PyMuPDF. This means:

• No PyMuPDF dependency: PyMuPDF will be removed from the requirements.txt file.
• Error handling: An informative error message will guide the user to install PyMuPDF if it's not already present in their environment. This error will only occur when PyMuPDF functionality is actually required.

Additionally, if this PR is accepted, the following actions will be necessary:

• Release a new version of PaddleOCR.
• Update relevant documentation within this repository.

I've used addle.util.try_import in the implementation, we can also discuss on this implementation.

PR 类型 PR types

Others

PR 变化内容类型 PR changes

Others

描述 Description

see above

提PR之前的检查 Check-list

• 这个 PR 是提交到dygraph分支或者是一个cherry-pick，否则请先提交到dygarph分支。
  This PR is pushed to the dygraph branch or cherry-picked from the dygraph branch. Otherwise, please push your changes to the dygraph branch.
• 这个PR清楚描述了功能，帮助评审能提升效率。This PR have fully described what it does such that reviewers can speedup.
• 这个PR已经经过本地测试。This PR can be covered by existing tests or locally verified.

[paddle-bot]

Thanks for your contribution!

[tink2123]

LGTM

[jzhang533]

will merge with an approval from @dyning

[dhruv-anand-aintech]

Thanks a lot for fixing this!

Do you know which version this will be available in?
I'm trying to install the latest version from pypi (2.7.0.3) and it still has it as a requirement, and fails due to the wheel error.

[jzhang533]

Thanks a lot for fixing this!

Do you know which version this will be available in?

I'm trying to install the latest version from pypi (2.7.0.3) and it still has it as a requirement, and fails due to the wheel error.

try this : https://pypi.org/project/paddleocr/2.7.2/

[dcfabian]

Hello,

I've noticed that PyMuPDF is still a dependency for PaddleOCR, albeit indirectly, and I wanted to bring this to attention in the context of this pull request. The dependency chain involves pdf2docx, which is listed as a requirement in ppstructure/recovery/requirement.txt.

Specifically, pdf2docx has PyMuPDF>=1.19.0 as a direct dependency. Since pdf2docx is required for PaddleOCR's ppstructure recovery functionality, it inherently makes PyMuPDF an indirect yet crucial dependency for the project.