Separate experimental scanner. Does not read or modify MicroManager inventory, suppliers, tasks or invoices. No invoice is uploaded to a server by this page's code; OCR libraries and models load from external CDNs.
For very small print, crop to the merchandise table or scan the top and bottom halves separately. This is a manual region crop, not automatic document perspective correction.
Waiting for image.
A second OCR pass on a selected table region, useful for tiny UNFI print. This preserves the original OCR results for comparison. It does not automatically approve or merge conflicting values.
Select the invoice photo above first.
Runs grayscale and adaptive-threshold passes on the same manually selected table region. Results stay separate. Use the outputs to find OCR disagreements; no fields are approved automatically. Two passes may be slow on iPhone.
Ready.
After selecting a photograph, set the table's top/bottom and the item-code column in the region controls above. This experimental tool slices that region into equal-height strips, OCRs each strip separately, and shows raw results. It does not assume every strip equals one product; row boundaries must be verified against the photograph. No inventory changes.
Compare item codes from two independently scanned OCR outputs. Agreement is evidence, not proof. Disagreements and missing values stay unresolved. This audit never writes inventory.
Import the JSON you already exported. No OCR rerun needed. Reference invoices are deliberately not bundled in the public deployment. Reconstruction works entirely offline after the page loads.
| # | Detected item code | Product text (unverified) | Qty/cost region (unverified) |
|---|
Compares both earlier generic reconstruction methods without hiding disagreements. A field is never accepted just because its text looks plausible. No inventory changes are possible.
Select an invoice fixture to compare detected item codes. This measures item-code recovery only; it does not certify descriptions, quantities, prices, or totals. No supplier-specific parser is used.
| Y | Raw text | Detected quantities / amounts | Flags |
|---|
Enter expected product row count from the original invoice. This measures candidate-row detection, not field accuracy. All recognized text remains available for inspection.