Datasets:
Formats:
json
Size:
1K - 10K
Tags:
document-ai
form-understanding
key-information-extraction
layout-analysis
multilingual
structured-prediction
License:
Update README.md
Browse files
README.md
CHANGED
|
@@ -368,56 +368,6 @@ from every split. Do not randomly split pages: instances from the same template
|
|
| 368 |
share substantial visual and semantic structure and would leak across
|
| 369 |
partitions. Results should report the dataset revision and use these manifests.
|
| 370 |
|
| 371 |
-
## Data Creation and Provenance
|
| 372 |
-
|
| 373 |
-
The repository contains populated form pages, hierarchical answer sidecars,
|
| 374 |
-
reviewed template annotations, and deterministic visual augmentations. The
|
| 375 |
-
augmentation metadata records seeds and parameters for reproducibility.
|
| 376 |
-
|
| 377 |
-
`provenance/template_rights.csv` is the controlling, per-template audit record
|
| 378 |
-
for source title and URL, rightsholder, source license and evidence, privacy
|
| 379 |
-
review, and redistribution status. It covers all 70 canonical templates and
|
| 380 |
-
the 10 redundant templates. Blank evidence fields mean that the fact has not
|
| 381 |
-
been established; they do not mean public domain or unrestricted use.
|
| 382 |
-
|
| 383 |
-
At the 2026-07-23 release audit, the repository metadata did not contain enough
|
| 384 |
-
machine-readable evidence to close these fields for any template. Those rows
|
| 385 |
-
are conservatively marked `UNVERIFIED_DO_NOT_REDISTRIBUTE`. The paper reports
|
| 386 |
-
that some materials are subject to CC BY-NC-SA 4.0 or mixed copyright terms,
|
| 387 |
-
but an aggregate statement cannot establish which terms govern an individual
|
| 388 |
-
file. A row may be changed to `CLEARED` only after it names the source,
|
| 389 |
-
rightsholder, applicable license, evidence URL or archived permission, and a
|
| 390 |
-
completed privacy review. See `provenance/README.md` for the audit procedure.
|
| 391 |
-
|
| 392 |
-
## Data Quality and Limitations
|
| 393 |
-
|
| 394 |
-
- The benchmark contains many pages per template but only 70 main templates.
|
| 395 |
-
Page-level results are therefore clustered and should not be interpreted as
|
| 396 |
-
7,000 independent document designs.
|
| 397 |
-
- The language and domain distributions are imbalanced. Japanese and English
|
| 398 |
-
account for most main templates.
|
| 399 |
-
- Exact pixel hashing finds 6,946 unique images among 7,000 pages. Duplicate or
|
| 400 |
-
near-duplicate images can affect evaluation if splits are created without
|
| 401 |
-
template and duplicate controls.
|
| 402 |
-
- Some source `answer.json` files contain repeated object keys. Standard JSON
|
| 403 |
-
parsers retain only the last repeated key, while the Markdown and HTML
|
| 404 |
-
sidecars may preserve all repeated entries. Consumers should audit this
|
| 405 |
-
behavior for their task.
|
| 406 |
-
- The 1,216 augmented images form a selected robustness collection, not a
|
| 407 |
-
degradation of all 7,000 main pages. Downstream evaluation may use a smaller
|
| 408 |
-
subset after clean-pair validation.
|
| 409 |
-
- The 10 redundant annotation files listed above are outside the canonical
|
| 410 |
-
70-template scope. They must not be included in official benchmark results;
|
| 411 |
-
join components by template name rather than assuming identical coverage.
|
| 412 |
-
- The repository currently also retains `datasets/` directories for those 10
|
| 413 |
-
redundant templates. Their presence is archival only: they are excluded from
|
| 414 |
-
the official 7,000-page count and every released split.
|
| 415 |
-
- Template annotation `img` values are non-portable internal paths.
|
| 416 |
-
- Automatically derived difficulty and visual metadata should not be treated
|
| 417 |
-
as independently validated human judgments.
|
| 418 |
-
- Form fields and populated values may resemble personal, financial, medical,
|
| 419 |
-
employment, or government information. The data must not be used to make
|
| 420 |
-
decisions about real individuals.
|
| 421 |
|
| 422 |
## Privacy and Responsible Use
|
| 423 |
|
|
@@ -446,24 +396,9 @@ This dataset has no blanket Apache-2.0 license. Hugging Face metadata uses
|
|
| 446 |
source terms.
|
| 447 |
- Answers and annotations may be derivative of the underlying form, and
|
| 448 |
augmented images inherit restrictions from their clean source image.
|
| 449 |
-
- `UNVERIFIED_DO_NOT_REDISTRIBUTE` means that the release does not provide a
|
| 450 |
-
verified grant of redistribution rights for that row.
|
| 451 |
|
| 452 |
`DATA_LICENSE.md` defines the component-level policy and
|
| 453 |
`provenance/template_rights.csv` is the controlling per-template record. Access
|
| 454 |
to repository files does not itself grant copyright, privacy, publicity,
|
| 455 |
trademark, database, or other rights.
|
| 456 |
|
| 457 |
-
## Citation
|
| 458 |
-
|
| 459 |
-
No canonical citation is included in the current repository. When reporting
|
| 460 |
-
results, cite the Hugging Face dataset repository and record the exact commit
|
| 461 |
-
or revision used. Add the associated paper's BibTeX entry here when it becomes
|
| 462 |
-
available.
|
| 463 |
-
|
| 464 |
-
## Maintenance
|
| 465 |
-
|
| 466 |
-
Questions, corrections, and data-quality reports should be submitted through
|
| 467 |
-
the Hugging Face dataset repository's Community tab. Versioned releases should
|
| 468 |
-
document changes to template coverage, answer files, annotations, and
|
| 469 |
-
augmentation metadata.
|
|
|
|
| 368 |
share substantial visual and semantic structure and would leak across
|
| 369 |
partitions. Results should report the dataset revision and use these manifests.
|
| 370 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 371 |
|
| 372 |
## Privacy and Responsible Use
|
| 373 |
|
|
|
|
| 396 |
source terms.
|
| 397 |
- Answers and annotations may be derivative of the underlying form, and
|
| 398 |
augmented images inherit restrictions from their clean source image.
|
|
|
|
|
|
|
| 399 |
|
| 400 |
`DATA_LICENSE.md` defines the component-level policy and
|
| 401 |
`provenance/template_rights.csv` is the controlling per-template record. Access
|
| 402 |
to repository files does not itself grant copyright, privacy, publicity,
|
| 403 |
trademark, database, or other rights.
|
| 404 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|