fix: handle return_word_box=True for Arabic text recognition

[brandenkmurray]

Summary

Fix TypeError when using Arabic OCR models with return_word_box=True.

Related issue: PaddlePaddle/PaddleOCR#16952

Root Cause

When return_word_box=True, CTCLabelDecode returns texts as a list of tuples (text_string, word_info_list) instead of plain strings (see processors.py line 294). The Arabic bidi reshaping code at predictor.py line 139 passes these tuples directly to get_display(), which expects a unicode string.

Fix

Check whether each element in texts is a tuple. If so, apply get_display() only to the text string while preserving the word box info. Plain string mode is unaffected.

Reproduction

from paddleocr import PaddleOCR

ocr = PaddleOCR(lang='ar')
result = ocr.predict('arabic_text.png', return_word_box=True)  # crashes without fix

[CLAassistant]

All committers have signed the CLA.

[paddle-bot]

Thanks for your contribution!

[luotao1]

Related issue: https://github.com/PaddlePaddle/PaddleX/issues/16952

404 error

[brandenkmurray]

Related issue: https://github.com/PaddlePaddle/PaddleX/issues/16952

404 error

Sorry, should be PaddlePaddle/PaddleOCR#16952