OCR Test PDF Sets

PDF files designed for OCR engine evaluation — each variant targets a specific recognition challenge. Ships with ground-truth text so you can compute character-level and word-level accuracy.

tables.pdf · MERGED CELLS · GROUND-TRUTH

01WHAT'S IN THE PACK

ocr-test-pack.zip contains:

  • clean-sans-serif.pdfHigh-contrast, single-column, 12pt sans-serif text. No ligatures, no kerning anomalies. Purpose: accuracy baseline.
  • tables.pdfMulti-column data tables with merged cells, numeric alignment, and thin rules. Purpose: table-structure recognition and cell-boundary detection.
  • two-column.pdfAcademic two-column layout with justified text and inline math symbols. Purpose: reading-order detection, column segmentation.
  • low-quality-scan.pdfSimulated 150 DPI scan: slight rotation (0.5°), uneven margins, toner-fade artifacts, JPEG compression noise. Purpose: robustness under degraded input.
  • ground-truth.txtUTF-8 text file with the exact expected content of each PDF, delimited by filename headers. Use for CER/WER computation.
  • README.mdPer-file notes: what each variant tests, recommended evaluation metrics.

02HOW TO USE

  1. 01
    Download and extract the ZIP.
  2. 02
    Run your OCR engine on each PDF.
  3. 03
    Compare output against the corresponding section in ground-truth.txt.
  4. 04
    Compute Character Error Rate (CER) or Word Error Rate (WER).
  5. 05
    Use the clean-sans-serif baseline to verify your pipeline works before testing harder variants.

03DOWNLOAD

Download ocr-test-pack.zip Generate a Scan-Style PDF → STATIC CDN · NO GENERATION DELAY · NO ACCOUNT

Character Error Rate (CER) and Word Error Rate (WER) against the ground-truth text. The clean-sans-serif file serves as an upper-bound baseline; degraded variants reveal where your engine loses accuracy.

It is a digitally rendered PDF that simulates common scan artifacts: slight rotation, uneven margins, toner fade, and JPEG compression. It is deterministic and reproducible, unlike a real scan.

→ All CollectionsBrowse every purpose-classified set. 5 SETS → GeneratorBuild a scan-style or table PDF. TOOL → RAG Test SetsDocument-type corpus for chunking tests. SET 01