hyperbots-inc/synthetic-doc-parsing

Synthetic Document Parsing

100 page images covering all 31 enterprise document types, with pixel-exact layout ground truth. Every page has its text blocks with labels and bounding boxes in reading order, every table as HTML, a Markdown rendering and the full OCR text. Use it to test the parse step that every downstream document task depends on.

It matches the output of HyperAPI POST /v1/parse, which returns labeled layout regions, reading order, tables and Markdown.

What's on each page

• Letterheads with a logo, title, key-value grids, section headers, paragraphs, bullet lists, multi-column tables, totals blocks, checkbox groups, signature lines and page footers.
• 12 print typefaces, 5 table styles (full grid, row rules, zebra, header-only, dark header) and random accent colours and margins.
• About 40% of the form pages are filled in handwriting-style script.

Block labels

page_header, title, logo, section_header, text, list, key_value, table, checkbox_group, signature, page_footer, page_number

Columns

Column	Description
image	1240 x 1754 JPEG (A4 at 150 DPI)
doc_type, taxonomy_id, taxonomy_name, category	Document type
blocks	List of {id, order, label, bbox, text}, where bbox = [x0, y0, x1, y1] in pixels and order = reading order
tables	List of {block_id, n_rows, n_cols, html}
markdown	Page as Markdown (headings, key-values, GitHub tables, task-list checkboxes)
ocr_text	Plain text of all blocks in reading order

from datasets import load_dataset
ds = load_dataset("hyperbots-inc/synthetic-doc-parsing", split="test")
ex = ds[0]
ex["image"]                      # PIL image
[b["label"] for b in ex["blocks"]]
print(ex["markdown"][:600])

Suggested metrics

Use text-level edit distance (CER/WER) for ocr_text, TEDS for tables[].html, mAP@IoU 0.5 for blocks by label, and Kendall's tau for reading order.

How the data was made

Every document is generated from a seeded random number generator by docgen.js and rendered on an HTML canvas in Chrome on macOS. Images use the system's print fonts and, for handwriting, the handwriting-style fonts installed with macOS (Bradley Hand, Noteworthy, Marker Felt, Chalkboard SE, Snell Roundhand, Apple Chancery, Savoye LET, Brush Script MT). Labels are taken from the generator itself, so there is no human annotation error. Field and block boxes are recorded at the moment each element is drawn.

All companies, people, addresses, account numbers and amounts are fictional. Phone numbers use ranges reserved for fiction where these exist (US 555-01xx, UK 07700 900xxx, AU 0491 570 1xx). Email addresses use the reserved example.com / example.org domains, and US ID numbers start with 9, which is never issued as an SSN.

To rebuild the dataset, load generator/docgen.js and generator/datasets.js in a browser page and run await DocSets.build('parsing'). Rendering depends on installed fonts, so pixels can differ on other systems, but text and labels are identical for the same seed.

Limitations

• The documents are synthetic. Their layouts, vocabulary and value ranges are much narrower than real enterprise traffic. Use them for benchmarking, regression tests and pipeline development, not as a replacement for evaluation on your own documents.
• All content is in English, with values formatted for 9 locales (US, UK, IN, CA, AU, DE, FR, SG, AE).
• Handwriting is rendered from fonts with per-character jitter. It is a stand-in for real handwriting, not a sample of it.

About HyperAPI

HyperAPI is Hyperbots' document intelligence API for enterprise documents. It provides parse, extract, classify, split and redact behind one endpoint and returns schema-validated JSON with a confidence score on every field. It has a REST API and a Python SDK, with an MCP server for agents available on request, and you can start free. See the HyperAPI Space for a walkthrough.

Part of the Hyperbots synthetic document suite

Dataset	Focus
synthetic-enterprise-doc-extraction	Field extraction across 31 enterprise document types
synthetic-doc-parsing (this dataset)	Layout blocks, reading order, Markdown and tables
synthetic-pii-redaction	PII spans and boxes for masking and de-identification
synthetic-doc-classification	31-class document classification
synthetic-doc-splitting	Multi-document bundles with true boundaries
synthetic-handwritten-forms	Forms filled in handwriting-style script
synthetic-stamped-documents	Stamps and seals over printed text
synthetic-bad-scans	Skew, blur, noise, shadows and compression at 3 severities
synthetic-dense-documents	Small type, multi-column layouts and long tables
synthetic-invoice-extraction	Invoice fields and line items (text + HTML)

License and citation

CC BY 4.0. Generator code is MIT.

@misc{hyperbots2026docparsing,
  title  = {Synthetic Document Parsing},
  author = {Hyperbots Inc.},
  year   = {2026},
  url    = {https://huggingface.co/datasets/hyperbots-inc/synthetic-doc-parsing}
}