30K synthetic invoices, receipts and bills from nine tax regimes with field-level bounding boxes.
Document models usually fail on the paperwork they have never seen: a German VAT invoice, a UAE tax invoice, an Indian GST bill, a US sales-tax receipt. This dataset generates all of them — nine tax regimes, four languages, realistic layouts, logos, stamps and handwriting — as clean scans, skewed scans and phone photographs.
Because every document is generated, the labels are pixel-perfect: each field and line item carries its bounding box, and totals are internally consistent, so you can validate extraction arithmetic as well as detection.
| Field | Type | Description |
|---|---|---|
document_id / file_name |
string | Identifier and path in the archive |
document_type |
string | invoice, receipt or utility_bill |
tax_regime |
string | VAT, GST, sales_tax and the issuing country |
currency / language |
string | Document currency and language |
capture |
string | clean_scan, skewed_scan or phone_photo |
fields |
array | Key-value pairs with bounding boxes |
line_items |
array | Description, quantity, unit price, tax rate, amount |
totals |
json | Net, tax and gross totals that reconcile |
split |
string | train, validation or test |
{
"document_id": "DOC-110245",
"file_name": "images/invoice/DOC-110245.png",
"document_type": "invoice",
"tax_regime": "VAT_DE",
"currency": "EUR",
"language": "de",
"capture": "phone_photo",
"fields": [
{
"key": "invoice_number",
"value": "RE-2026-0412",
"bbox": [
812,
144,
1030,
176
]
},
{
"key": "vat_id",
"value": "DE123456789",
"bbox": [
96,
232,
318,
262
]
},
{
"key": "total_gross",
"value": "1.860,00",
"bbox": [
901,
1422,
1068,
1458
]
}
],
"totals": {
"net": 1563.03,
"tax_rate": 19,
"tax": 296.97,
"gross": 1860
}
}
Illustrative sample showing the structure of the data. The full dataset is delivered after purchase.
42K shelf photographs with 1.3M product boxes, out-of-stock gaps and planogram compliance labels.
Half-hourly stock, delivery-time and price snapshots from 900 dark stores in 14 cities.
24 months of matched prices, sales rank and review velocity for 120K SKUs across six marketplaces.
Tell us what you are trying to predict, automate or understand. We will reply within one business day with an approach, a timeline and a quote.