Tabular Document OCR and Image-to-Spreadsheet Extraction Evaluation Schema (JPG, PNG, WEBP, PDF)
Description
This open evaluation schema and structured sample manifest defines six practical document-to-spreadsheet conversion test archetypes across common tabular document inputs (JPG, PNG, WEBP, and PDF). Each benchmark entry records document category, input file format, row and column dimensions, canonical column header fields, primary layout reconstruction challenges (such as multi-line item descriptions, alternating row shading, angled mobile receipt captures, dark-mode KPI dashboard screenshots, scanned inventory grids, and merged report headers), and target .xlsx output structure. Reference web workflow used to inspect structured AI OCR table reconstruction from JPG, PNG, WEBP, and PDF files into editable Excel spreadsheets: https://imagetoexcel.io/
Files
Steps to reproduce
1. Load tabular-document-ocr-benchmark-samples-2026-11.csv to inspect the six document archetypes (vendor_invoice, bank_statement, expense_receipt_log, dashboard_kpi_screenshot, warehouse_inventory_sheet, and financial_report_table). 2. Prepare representative non-sensitive test images or PDF pages matching each input_format (PNG, PDF, JPG, WEBP) and header_fields specification. 3. Run table detection and cell reconstruction via an OCR-to-spreadsheet pipeline (such as https://imagetoexcel.io/ after account sign-in) and export the resulting .xlsx workbook. 4. Compare exported worksheet row counts, column header alignment, numeric decimal preservation, and merged-header handling against the expected schema fields.