jedarden/pdftract
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
pdftract/notes/bf-204mm.md
jedarden bc9e6f8e57 test(bf-1b7od): add profile_yaml fuzz target and verify cargo-fuzz configuration 
- Create fuzz/fuzz_targets/profile_yaml.rs to test YAML parsing
- Register profile_yaml target in fuzz/Cargo.toml
- Verify cargo-fuzz 0.13.1 is installed and working
- All 7 fuzz targets now operational per plan requirement (line 3236)

Closes bf-1b7od. Verification: notes/bf-1b7od.md
2026-07-06 09:38:59 -04:00

3.4 KiB
Raw Blame History

bf-204mm: Multi-Page Report OCR Fixture

Task

Add multi-page report OCR fixture (≥5 pages) with ground truth text for WER testing.

What Was Done

Files Modified

  • tests/fixtures/scanned/multi-page/report-300dpi.pdf - Replaced text-embedded version with scanned version (2.2M)
  • tests/fixtures/scanned/multi-page/report-300dpi-ground-truth.txt - Renamed from report-300dpi.txt

Implementation Notes

The fixture already existed as doc-10page-300dpi-scanned.pdf with identical ground truth content. The task requirement was to have it named as report-300dpi.pdf with proper -ground-truth.txt suffix.

Source

  • Content: Public-domain multi-page document (government report-style content)
  • Original fixture: doc-10page-300dpi-scanned.pdf (same content)

Acceptance Criteria - PASS

✅ PASS: report-300dpi.pdf exists with ≥5 pages

  • Verified: 11 pages
  • Command: pdfinfo tests/fixtures/scanned/multi-page/report-300dpi.pdf | grep Pages
  • Output: Pages: 11

✅ PASS: Ground truth .txt file exists

  • File: tests/fixtures/scanned/multi-page/report-300dpi-ground-truth.txt
  • Size: 12K
  • Content: Complete text transcript of all 11 pages

✅ PASS: Running scripts/measure-wer.sh with perfect OCR gives WER = 0%

  • Command: ./scripts/measure-wer.sh tests/fixtures/scanned/multi-page/report-300dpi-ground-truth.txt tests/fixtures/scanned/multi-page/report-300dpi-ground-truth.txt
  • Output: WER: 0.0000 (0.00%)
  • This confirms the WER calculation works correctly when OCR output matches ground truth exactly

✅ PASS: PDF is ~300 DPI

  • Image dimensions: 2550 x 3300 pixels per page
  • Page size: Letter (8.5 x 11 inches)
  • DPI calculation: 2550 / 8.5 = 300 DPI, 3300 / 11 = 300 DPI
  • Verified via: pdfimages -list showing 2550x3300 RGB images

Technical Details

Fixture Properties

  • Format: Scanned PDF (image-based, no embedded text)
  • Pages: 11 pages
  • Resolution: 300 DPI (2550x3300 pixels per page on Letter size)
  • Color: RGB
  • File size: 2.2M
  • Content types: Introduction, text-heavy content, forms, tables, technical specs, legal text, financial statements, correspondence, scientific content, summary

Content Variety

The fixture contains diverse content types for comprehensive OCR testing:

  1. Introduction and overview
  2. Text-heavy technical documentation
  3. Form with fields and checkboxes
  4. Tabular data with formatting
  5. API technical specifications
  6. Legal terms and conditions
  7. Financial balance sheet
  8. Business correspondence
  9. Scientific abstract and methodology
  10. Summary and conclusions

Verification Steps Completed

  1. ✅ Verified page count (11 pages ≥ 5 required)
  2. ✅ Verified ground truth file exists with correct naming
  3. ✅ Verified WER calculation returns 0% for perfect match
  4. ✅ Verified PDF is scanned (images) not text-embedded
  5. ✅ Verified DPI is ~300 (300 x 300)

Commit Information

  • Files changed: 2 files (renamed/copied)
  • Commit message: feat(bf-204mm): add multi-page report OCR fixture

Notes

  • The fixture content is identical to doc-10page-300dpi-scanned.pdf (same MD5 checksum for ground truth)
  • This is intentional - the task required the report-300dpi.pdf naming convention
  • The fixture supports OCR performance testing across page boundaries and longer documents
  • Content sourced from public-domain test document generation (not a real government report, but realistic content)