Goodspotai/goodspot-eu-na-receipts

Goodspot EU/NA Receipts — Labels

High-quality labels for a 320-receipt evaluation set drawn from European and North American receipts. Used in the Goodspot project to benchmark receipt-extraction models (LFM2-VL, fine-tuned variants, baselines).

This repo contains labels only. The receipt images are not redistributed here.

Files

• labels_hq.jsonl — one JSON object per line, 320 records.

Fields

Field	Type	Description
id	string	Stable record id (e.g. eu_na_0025).
image_path	string	Relative path to the receipt image in the upstream project layout.
merchant	string | null	Store / merchant name as printed on the receipt.
date	string | null	Transaction date, YYYY-MM-DD.
total	float | null	Receipt total as a positive float, no currency symbol.
category	string	Coarse spend category (e.g. Eating Out, Groceries).
source_pdf	string	Internal source path the receipt was extracted from.
ocr_text_available	bool	Whether OCR text is available for the corresponding image.

Example record

{
  "id": "eu_na_0025",
  "image_path": "eu_na_receipts/img/cafecrepe_20180908_008_0025.jpg",
  "merchant": "Cafe Crepe",
  "date": "2018-08-31",
  "total": 13.13,
  "category": "Eating Out",
  "source_pdf": "receipts_raw/2018/ca/restaurant/cafecrepe_20180908_008.pdf",
  "ocr_text_available": true
}

Loading

from datasets import load_dataset
ds = load_dataset("Goodspotai/goodspot-eu-na-receipts", data_files="labels_hq.jsonl", split="train")
print(ds[0])

Or with plain Python:

import json
with open("labels_hq.jsonl") as f:
    records = [json.loads(line) for line in f]

Intended use

• Benchmarking receipt-extraction models on the four-field task (merchant, date, address, total).
• Scoring with the matcher in backend/scripts/eval_helpers.py (exact-match on normalised text, abs(gt - pr) < 0.01 tolerance on totals).

Out-of-scope

• Training a production model on this set alone — it is a hand-verified evaluation slice, not a training corpus.
• Any task requiring the receipt images themselves; only labels are released here.

Provenance

Labels were extracted from PDF receipts originating in EU and NA jurisdictions and then hand-verified against the rendered receipt images. The verification workflow normalises dates to YYYY-MM-DD and totals to positive floats with no currency symbol.

License

Released under CC BY 4.0. Please cite the Goodspot project if you use this set in published work.