Create Datasets

import asyncio
from penquify.models import Document, DocHeader, DocItem
from penquify.generators.verify import generate_verified_dataset

doc = Document(
    header=DocHeader(
        doc_type="guia_despacho",
        doc_number="00054321",
        date="15/04/2026",
        emitter_name="ACME FOODS S.A.",
        receiver_name="DISTRIBUIDORA CENTRAL LTDA.",
    ),
    items=[
        DocItem(pos=1, code="AF-1001", description="HARINA DE TRIGO PREMIUM 25KG",
                qty=40, unit="UN", unit_price=12500, total=500000),
        DocItem(pos=2, code="AF-2003", description="ACEITE VEGETAL 5L",
                qty=24, unit="UN", unit_price=6990, total=167760),
    ],
)

results = asyncio.run(generate_verified_dataset(
    reference_image_path="output/guia_despacho_00054321.png",
    document=doc,
    output_dir="output/dataset",
    preset_names=["full_picture", "folded_skewed", "blurry", "coffee_stain"],
    max_retries=3,
))

output/dataset/
  ground_truth.json                     # master ground truth (all fields)
  dataset_summary.json                  # summary: total, verified, failed
  photo_full_picture.png
  photo_full_picture_verification.json
  photo_full_picture_occlusion.json
  photo_folded_skewed.png
  photo_folded_skewed_verification.json
  photo_folded_skewed_occlusion.json
  ...

{
  "total": 4,
  "verified": 3,
  "failed": 1,
  "results": [
    {
      "name": "photo_full_picture.png",
      "verified": true,
      "attempts": 1,
      "summary": {
        "total_fields": 15,
        "matched": 15,
        "mismatched": 0,
        "illegible": 0,
        "not_visible": 0
      }
    }
  ]
}

import json
import asyncio
from penquify.models import Document, DocHeader, DocItem
from penquify.generators.pdf import generate_document_files
from penquify.generators.verify import generate_verified_dataset

async def build_benchmark(docs_json_path: str, output_base: str):
    with open(docs_json_path) as f:
        docs_data = json.load(f)

    presets = ["full_picture", "folded_skewed", "blurry",
               "cropped_header", "coffee_stain", "strong_oblique"]

    for doc_data in docs_data:
        doc = Document(
            header=DocHeader(**doc_data["header"]),
            items=[DocItem(**it) for it in doc_data["items"]],
        )
        doc_dir = f"{output_base}/{doc.header.doc_number}"

        # Generate clean document
        files = await generate_document_files(doc, doc_dir)

        # Generate verified dataset
        await generate_verified_dataset(
            reference_image_path=files["png"],
            document=doc,
            output_dir=f"{doc_dir}/photos",
            preset_names=presets,
            max_retries=3,
        )
        print(f"Done: {doc.header.doc_number}")

asyncio.run(build_benchmark("my_docs.json", "benchmark/"))

  [OK] full_picture: output/dataset/photo_full_picture.png
  [VERIFIED] full_picture: attempt 1, 15 match, 0 illegible, 0 not_visible
  [RETRY 1/3] folded_skewed: 2 mismatched fields: ['item_1_total', 'total']
  [VERIFIED] folded_skewed: attempt 2, 13 match, 2 illegible, 0 not_visible
  [FAIL] blurry: 1 mismatches after 3 attempts

Dataset: 3/4 verified, 1 failed