[Yi-111-a]

Summary

_collect_annotation_content stopped after the first non-empty line for any label other than tables and pictures. A DeepSeek-OCR response like

text[[100,100,900,300]]
First line
Second line

text[[100,400,900,500]]
Next block

produced First line and Next block as text items and silently dropped Second line. Unlimited-OCR delegates to the same collector after normalizing its <|det|>.. markers, so it lost the same lines.

The collector now gathers the block's consecutive non-empty lines until the next annotation or the first blank line. Empty lines, following annotations, captions, and table markup keep their boundaries, so caption linkage and table parsing are unchanged; figure/image/table blocks are handled by their existing branches. This fixes both parsers.

Closes #4905

Testing

• New plain-string tests in tests/test_unlimited_ocr_markdown.py pinning the issue's reproduction for both response shapes: continuation lines survive, and an annotation line / blank line still ends a block.
• uv run pytest -q tests/test_unlimited_ocr_markdown.py: 9 passed.
• tests/test_deepseekocr_vlm.py cannot be collected locally (it imports the full VLM pipeline stack, which needs the models extra); its reference-data paths are untouched by this change.

[github-actions]

✅ DCO Check Passed

Thanks @Yi-111-a, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🟢 Merge protection satisfied — ready to merge.

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[github-actions]

AI triage (advisory)

Checked commit 1f6aacf744. The analysis read the diff only. It did not run code from this pull request.

This pull request updates _collect_annotation_content to collect consecutive non-empty lines for text blocks in DeepSeek-OCR and Unlimited-OCR. It also adds unit tests for multiline OCR text blocks.

CI safety: no concern found. A maintainer must still approve CI.

Possible duplicates

PR	State	Signal	Assessment
#4909	open	also references issue #4905; changes the same code	duplicate: PR #4909 changes the same function to preserve multiline text blocks for DeepSeek-OCR and Unlimited-OCR.

_{The ai:* labels are advisory. They do not approve CI and they do not add tests:full. A new push runs this check again.}

[github-actions]

AI first review (advisory): no blocker found

This review covers the full diff at 8523cc67a3. It read the code and did not run it.

No blocking defect found. The new loop in _collect_annotation_content stops at the next annotation or the first blank line after content. That keeps the old boundaries for tables, captions and figures. The three DeepSeek reference sources in tests/data/md_deepseek/sources put each block's content on one line before a blank line, so their output does not change.

_{A maintainer makes the merge decision. This review never approves the pull request.}

[Yi-111-a]

DCO check fixed: the single commit now carries the Signed-off-by trailer (amended 8523cc6). No content change.

[dolfim-ibm]

lgtm

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[cau-git]

@Yi-111-a this has been already handled in #4964, thanks. I am therefore closing this PR.