Request to Add return_word_box Argument in Pip Release of PaddleOCR

[thainv02]

Hi PaddleOCR team,

First of all, thank you for the amazing work on PaddleOCR. It's been a very useful tool for my OCR-based projects.

🚀 Feature Request
I would like to request that the return_word_box argument be officially supported in the pip-installed version of PaddleOCR (i.e., via pip install paddleocr). Currently, this argument is available in the latest source code on GitHub, but it is not usable directly from the pip package.

🎯 Why this is needed
I'm working on a multi-language OCR system (Vietnamese, English, Korean) and I need to retrieve the bounding boxes of individual words, not just lines. The return_word_box=True argument works well when using the latest development code (cloned from GitHub), but is missing in the pip version.

This feature would help:

Improve flexibility in downstream processing

Make the pip version feature-complete with the GitHub version

Simplify usage without needing manual installation from source

✅ Expected Behavior
Allow users to pass return_word_box=True when using the PaddleOCR Python API like:
ocr = PaddleOCR(use_angle_cls=True, lang="en", return_word_box=True)
And return results including individual word boxes, similar to the GitHub version.

📦 Current Workaround
I'm currently cloning the repo and using the code directly to get this feature. But this adds complexity and conflicts with virtual environments that rely on pip installation.

🙏 Request
Please consider including this feature in the next pip release. Even a temporary backport or workaround in the meantime would be appreciated.

Thanks again for all your contributions!

[zhangyubo0722]

Thank you very much for your recognition. This feature is expected to be supported in PaddleOCR version 3.2.