fix(table): escape html-sensitive OCR text in table markdown output

[jimmyzhuu]

Summary

This PR fixes HTML escaping for OCR text inserted into table cell markdown / HTML output.

It applies the fix to both table post-processing paths:

• table_recognition_post_processing.py
• table_recognition_post_processing_v2.py

The fix escapes HTML-sensitive OCR text before inserting it into generated <td> content, while preserving a single outer <b>...</b> wrapper when present.

Why This Change

When OCR text inside a table cell contains content such as <recv .../> or <pause .../>, the current HTML assembly path may inject raw OCR text directly into the generated table structure.

This breaks exported markdown / HTML rendering and mixes original OCR content with generated HTML.

The root cause is in PaddleX table post-processing rather than the PaddleOCR wrapper layer.

What Changed

• escape HTML-sensitive OCR text with html.escape(..., quote=True)
• preserve a single outer <b>...</b> wrapper for bold cell content
• keep the existing multi-fragment cell concatenation behavior
• apply the same behavior to both legacy and v2 table post-processing paths
• add regression tests for single-fragment, bold-wrapped, and multi-fragment OCR text

Tests

python -m pytest tests/unit/inference/pipelines/table_recognition/test_table_markdown_escaping.py -q
python -m py_compile \
  paddlex/inference/pipelines/table_recognition/table_recognition_post_processing.py \
  paddlex/inference/pipelines/table_recognition/table_recognition_post_processing_v2.py \
  tests/unit/inference/pipelines/table_recognition/test_table_markdown_escaping.py

Related

• root user report: PaddleOCR#16096
• migrated from: PaddleOCR#17924

[paddle-bot]

Thanks for your contribution!