File size: 23,315 Bytes
1ef1de7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 | """Public benchmark manifest readiness checks."""
from __future__ import annotations
import html
import json
import re
import shutil
from dataclasses import dataclass, field
from html.parser import HTMLParser
from pathlib import Path
from typing import Any
from .datasets import entry_slices
from .errors import DataValidationError
from .manifest import ManifestEntry, load_manifest, sha256_file, sha256_text, write_manifest
from .normalization import normalize_ocr_text
from .schemas import TableCell
@dataclass(frozen=True, slots=True)
class PublicBenchmarkSpec:
dataset: str
title: str
url: str
min_samples: int
required_slices: tuple[str, ...] = ()
notes: str = ""
def to_dict(self) -> dict[str, Any]:
return {
"dataset": self.dataset,
"title": self.title,
"url": self.url,
"min_samples": self.min_samples,
"required_slices": list(self.required_slices),
"notes": self.notes,
}
PUBLIC_BENCHMARKS: dict[str, PublicBenchmarkSpec] = {
"opendatalab/OmniDocBench": PublicBenchmarkSpec(
dataset="opendatalab/OmniDocBench",
title="OmniDocBench",
url="https://huggingface.co/datasets/opendatalab/OmniDocBench",
min_samples=100,
required_slices=("public", "document_parsing"),
notes="Real-world document parsing benchmark; use official annotations/scorer when available.",
),
"allenai/olmOCR-bench": PublicBenchmarkSpec(
dataset="allenai/olmOCR-bench",
title="olmOCR-bench",
url="https://huggingface.co/datasets/allenai/olmOCR-bench",
min_samples=100,
required_slices=("public", "pdf", "markdown"),
notes="PDF-to-Markdown benchmark with unit-test style assertions.",
),
"PaddlePaddle/Real5-OmniDocBench": PublicBenchmarkSpec(
dataset="PaddlePaddle/Real5-OmniDocBench",
title="Real5-OmniDocBench",
url="https://huggingface.co/datasets/PaddlePaddle/Real5-OmniDocBench",
min_samples=100,
required_slices=("public", "real_world", "document_parsing"),
notes="Physical reconstruction benchmark with scanning, warping, screen-photo, illumination, and skew scenarios.",
),
}
@dataclass(slots=True)
class PublicBenchmarkAudit:
manifest_path: str
passed: bool
benchmark: str
sample_count: int
dataset_counts: dict[str, int]
slice_counts: dict[str, int]
spec: dict[str, Any] | None = None
issues: list[str] = field(default_factory=list)
warnings: list[str] = field(default_factory=list)
summary_json_path: str | None = None
summary_md_path: str | None = None
def to_dict(self) -> dict[str, Any]:
return {
"manifest_path": self.manifest_path,
"passed": self.passed,
"benchmark": self.benchmark,
"sample_count": self.sample_count,
"dataset_counts": self.dataset_counts,
"slice_counts": self.slice_counts,
"spec": self.spec,
"issues": self.issues,
"warnings": self.warnings,
"summary_json_path": self.summary_json_path,
"summary_md_path": self.summary_md_path,
}
@dataclass(slots=True)
class PublicBenchmarkPrepareReport:
benchmark: str
manifest_path: str
references_dir: str
sample_count: int
filters: dict[str, str | int | bool | None]
summary_json_path: str
summary_md_path: str
def to_dict(self) -> dict[str, Any]:
return {
"benchmark": self.benchmark,
"manifest_path": self.manifest_path,
"references_dir": self.references_dir,
"sample_count": self.sample_count,
"filters": self.filters,
"summary_json_path": self.summary_json_path,
"summary_md_path": self.summary_md_path,
}
def list_public_benchmarks() -> list[PublicBenchmarkSpec]:
return [PUBLIC_BENCHMARKS[key] for key in sorted(PUBLIC_BENCHMARKS)]
def prepare_public_benchmark(
benchmark: str,
output_dir: str | Path,
*,
limit: int | None = None,
language: str | None = None,
subset: str | None = None,
data_source: str | None = None,
require_tables: bool = False,
annotation_path: str | Path | None = None,
local_image_root: str | Path | None = None,
) -> PublicBenchmarkPrepareReport:
if benchmark != "opendatalab/OmniDocBench":
raise DataValidationError(f"unsupported public benchmark converter: {benchmark}")
return _prepare_omnidocbench(
output_dir,
limit=limit,
language=language,
subset=subset,
data_source=data_source,
require_tables=require_tables,
annotation_path=annotation_path,
local_image_root=local_image_root,
)
def audit_public_benchmark_manifest(
manifest_path: str | Path,
output_dir: str | Path | None = None,
*,
benchmark: str | None = None,
min_samples: int | None = None,
require_reference_paths: bool = True,
) -> PublicBenchmarkAudit:
path = Path(manifest_path)
entries = load_manifest(path)
if not entries:
raise DataValidationError("public benchmark audit requires a non-empty manifest")
benchmark_name = benchmark or _infer_benchmark_name(entries)
spec = PUBLIC_BENCHMARKS.get(benchmark_name)
required_count = min_samples if min_samples is not None else (spec.min_samples if spec else 1)
issues: list[str] = []
warnings: list[str] = []
dataset_counts: dict[str, int] = {}
slice_counts: dict[str, int] = {}
for entry in entries:
dataset_counts[entry.dataset] = dataset_counts.get(entry.dataset, 0) + 1
for slice_name in entry_slices(entry):
slice_counts[slice_name] = slice_counts.get(slice_name, 0) + 1
metadata = entry.metadata or {}
if metadata.get("public_benchmark") is not True:
issues.append(f"{entry.sample_id}: metadata.public_benchmark must be true")
if not metadata.get("source_url"):
issues.append(f"{entry.sample_id}: metadata.source_url is required")
if not metadata.get("benchmark_name"):
issues.append(f"{entry.sample_id}: metadata.benchmark_name is required")
if require_reference_paths:
reference_path = metadata.get("reference_path")
if not isinstance(reference_path, str) or not reference_path:
issues.append(f"{entry.sample_id}: metadata.reference_path is required")
elif not Path(reference_path).exists():
issues.append(f"{entry.sample_id}: reference_path does not exist: {reference_path}")
if len(entries) < required_count:
issues.append(f"sample_count {len(entries)} is below required minimum {required_count}")
if spec is None:
warnings.append(f"benchmark {benchmark_name!r} is not in the public benchmark registry")
else:
for required_slice in spec.required_slices:
if slice_counts.get(required_slice, 0) == 0:
issues.append(f"required public benchmark slice missing: {required_slice}")
audit = PublicBenchmarkAudit(
manifest_path=str(path),
passed=not issues,
benchmark=benchmark_name,
sample_count=len(entries),
dataset_counts=dict(sorted(dataset_counts.items())),
slice_counts=dict(sorted(slice_counts.items())),
spec=spec.to_dict() if spec else None,
issues=issues,
warnings=warnings,
)
if output_dir is not None:
_write_public_benchmark_audit(audit, Path(output_dir))
return audit
def _prepare_omnidocbench(
output_dir: str | Path,
*,
limit: int | None,
language: str | None,
subset: str | None,
data_source: str | None,
require_tables: bool,
annotation_path: str | Path | None,
local_image_root: str | Path | None,
) -> PublicBenchmarkPrepareReport:
if limit is not None and limit < 1:
raise DataValidationError("limit must be positive when provided")
benchmark = "opendatalab/OmniDocBench"
out = Path(output_dir)
manifest_dir = out / "manifests"
image_dir = out / "images"
reference_dir = out / "references"
manifest_dir.mkdir(parents=True, exist_ok=True)
image_dir.mkdir(parents=True, exist_ok=True)
reference_dir.mkdir(parents=True, exist_ok=True)
rows, repo_files, download_file = _load_omnidocbench_source(annotation_path, local_image_root)
entries: list[ManifestEntry] = []
for row_index, row in enumerate(rows):
if limit is not None and len(entries) >= limit:
break
if not isinstance(row, dict):
raise DataValidationError(f"OmniDocBench row {row_index} must be an object")
page_info = row.get("page_info")
if not isinstance(page_info, dict):
raise DataValidationError(f"OmniDocBench row {row_index} is missing page_info")
attributes = page_info.get("page_attribute")
if not isinstance(attributes, dict):
raise DataValidationError(f"OmniDocBench row {row_index} is missing page_info.page_attribute")
if language and str(attributes.get("language")) != language:
continue
if subset and str(attributes.get("subset")) != subset:
continue
if data_source and str(attributes.get("data_source")) != data_source:
continue
layout_dets = row.get("layout_dets")
if not isinstance(layout_dets, list):
raise DataValidationError(f"OmniDocBench row {row_index} is missing layout_dets")
tables = _omnidocbench_reference_tables(layout_dets)
if require_tables and not tables:
continue
image_name = page_info.get("image_path")
if not isinstance(image_name, str) or not image_name:
raise DataValidationError(f"OmniDocBench row {row_index} page_info.image_path must be a non-empty string")
dataset_image_path = image_name if image_name in repo_files else f"images/{image_name}"
if dataset_image_path not in repo_files:
raise DataValidationError(f"OmniDocBench row {row_index} image file is missing from dataset repo: {dataset_image_path}")
source_image = download_file(dataset_image_path)
target_image = image_dir / Path(dataset_image_path).name
if source_image.resolve() != target_image.resolve():
shutil.copy2(source_image, target_image)
sample_id = f"omnidocbench-{row_index:05d}-{Path(image_name).stem}"
reference_path = reference_dir / f"{sample_id}.ref.json"
reference_text = _omnidocbench_reference_text(layout_dets)
reference_payload = {
"text": reference_text,
"tables": tables,
"metadata": {
"benchmark_name": benchmark,
"source_image_path": dataset_image_path,
"page_info": page_info,
},
}
reference_path.write_text(json.dumps(reference_payload, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
image_hash = sha256_file(target_image)
text_hash = sha256_text(reference_text)
slices = _omnidocbench_slices(attributes, tables=tables, layout_dets=layout_dets)
entries.append(
ManifestEntry(
sample_id=sample_id,
dataset=benchmark,
split="eval",
image_path=str(target_image),
text=reference_text,
sha256=image_hash,
metadata={
"benchmark_name": benchmark,
"public_benchmark": True,
"source_url": PUBLIC_BENCHMARKS[benchmark].url,
"source_image_path": dataset_image_path,
"reference_path": str(reference_path),
"slices": slices,
"text_sha256": text_hash,
"sample_sha256": sha256_text(f"{image_hash}\n{text_hash}"),
"page_attribute": attributes,
"page_no": page_info.get("page_no"),
"width": page_info.get("width"),
"height": page_info.get("height"),
},
)
)
if not entries:
raise DataValidationError("No OmniDocBench rows matched the requested filters")
manifest_path = manifest_dir / "omnidocbench-eval.jsonl"
write_manifest(entries, manifest_path)
report = PublicBenchmarkPrepareReport(
benchmark=benchmark,
manifest_path=str(manifest_path),
references_dir=str(reference_dir),
sample_count=len(entries),
filters={
"limit": limit,
"language": language,
"subset": subset,
"data_source": data_source,
"require_tables": require_tables,
},
summary_json_path=str(out / "public-benchmark-prepare.json"),
summary_md_path=str(out / "public-benchmark-prepare.md"),
)
_write_public_benchmark_prepare_report(report)
return report
def _load_omnidocbench_source(
annotation_path: str | Path | None,
local_image_root: str | Path | None,
) -> tuple[list[dict[str, Any]], set[str], Any]:
if annotation_path is not None:
annotation = Path(annotation_path)
if not annotation.exists():
raise DataValidationError(f"OmniDocBench annotation path does not exist: {annotation}")
root = Path(local_image_root) if local_image_root is not None else annotation.parent
files = {str(path.relative_to(root)) for path in root.rglob("*") if path.is_file()}
def local_download(name: str) -> Path:
path = root / name
if not path.exists():
raise DataValidationError(f"OmniDocBench local image file missing: {path}")
return path
rows = json.loads(annotation.read_text(encoding="utf-8"))
if not isinstance(rows, list):
raise DataValidationError("OmniDocBench annotation JSON must be a list")
return rows, files, local_download
try:
from huggingface_hub import hf_hub_download, list_repo_files
except ImportError as exc:
raise DataValidationError("OmniDocBench preparation requires huggingface_hub. Install ocr-tech[datasets].") from exc
benchmark = "opendatalab/OmniDocBench"
try:
annotation = Path(hf_hub_download(repo_id=benchmark, repo_type="dataset", filename="OmniDocBench.json"))
files = set(list_repo_files(repo_id=benchmark, repo_type="dataset"))
except Exception as exc:
raise DataValidationError(f"Failed to inspect OmniDocBench dataset repo: {exc}") from exc
def hf_download(name: str) -> Path:
try:
return Path(hf_hub_download(repo_id=benchmark, repo_type="dataset", filename=name))
except Exception as exc:
raise DataValidationError(f"Failed to download OmniDocBench file {name!r}: {exc}") from exc
rows = json.loads(annotation.read_text(encoding="utf-8"))
if not isinstance(rows, list):
raise DataValidationError("OmniDocBench annotation JSON must be a list")
return rows, files, hf_download
def _omnidocbench_reference_text(layout_dets: list[Any]) -> str:
parts: list[str] = []
for item in sorted(_valid_layout_items(layout_dets), key=lambda value: int(value.get("order", 0) or 0)):
category = str(item.get("category_type") or "")
if category == "table" and isinstance(item.get("html"), str):
text = _html_to_text(str(item["html"]))
else:
text = str(item.get("text") or "")
text = normalize_ocr_text(text, collapse_spaces=True).strip()
if text:
parts.append(text)
return "\n".join(parts)
def _omnidocbench_reference_tables(layout_dets: list[Any]) -> list[dict[str, Any]]:
tables: list[dict[str, Any]] = []
for item in sorted(_valid_layout_items(layout_dets), key=lambda value: int(value.get("order", 0) or 0)):
if item.get("category_type") != "table" or not isinstance(item.get("html"), str):
continue
cells = _table_cells_from_html(str(item["html"]))
if cells:
tables.append(
{
"cells": [cell.to_dict() for cell in cells],
"metadata": {"anno_id": item.get("anno_id"), "order": item.get("order")},
}
)
return tables
def _valid_layout_items(layout_dets: list[Any]) -> list[dict[str, Any]]:
return [item for item in layout_dets if isinstance(item, dict) and not bool(item.get("ignore", False))]
def _omnidocbench_slices(attributes: dict[str, Any], *, tables: list[dict[str, Any]], layout_dets: list[Any]) -> list[str]:
slices = ["public", "document_parsing"]
language = str(attributes.get("language") or "")
if language:
slices.append(language)
if language == "english":
slices.append("english")
if "chinese" in language:
slices.append("chinese")
subset = str(attributes.get("subset") or "")
if subset:
slices.append(subset)
source = str(attributes.get("data_source") or "")
if source:
slices.append(source)
categories = {str(item.get("category_type") or "") for item in _valid_layout_items(layout_dets)}
if tables:
slices.append("table")
if any("figure" in category for category in categories):
slices.append("figure")
if any("equation" in category for category in categories):
slices.append("equation")
return sorted(set(slices))
class _PlainHtmlTextExtractor(HTMLParser):
def __init__(self) -> None:
super().__init__()
self.parts: list[str] = []
def handle_starttag(self, tag: str, attrs: list[tuple[str, str | None]]) -> None:
if tag in {"tr", "br", "p", "div"}:
self.parts.append("\n")
if tag in {"td", "th"}:
self.parts.append(" ")
def handle_data(self, data: str) -> None:
self.parts.append(data)
def text(self) -> str:
value = re.sub(r"\n{3,}", "\n\n", re.sub(r"[ \t]+", " ", html.unescape("".join(self.parts)))).strip()
return "\n".join(line.strip() for line in value.splitlines() if line.strip())
def _html_to_text(value: str) -> str:
parser = _PlainHtmlTextExtractor()
parser.feed(value)
parser.close()
return parser.text()
class _TableHtmlParser(HTMLParser):
def __init__(self) -> None:
super().__init__()
self.cells: list[TableCell] = []
self._row = -1
self._col = 0
self._occupied: set[tuple[int, int]] = set()
self._cell_text: list[str] | None = None
self._rowspan = 1
self._colspan = 1
def handle_starttag(self, tag: str, attrs: list[tuple[str, str | None]]) -> None:
if tag == "tr":
self._row += 1
self._col = 0
return
if tag not in {"td", "th"} or self._row < 0:
return
while (self._row, self._col) in self._occupied:
self._col += 1
attrs_dict = {key.lower(): value for key, value in attrs}
self._rowspan = _positive_int(attrs_dict.get("rowspan"), default=1)
self._colspan = _positive_int(attrs_dict.get("colspan"), default=1)
self._cell_text = []
def handle_endtag(self, tag: str) -> None:
if tag not in {"td", "th"} or self._cell_text is None:
return
text = normalize_ocr_text(html.unescape("".join(self._cell_text)), collapse_spaces=True).strip()
self.cells.append(TableCell(row=self._row, col=self._col, text=text, rowspan=self._rowspan, colspan=self._colspan))
for row_offset in range(self._rowspan):
for col_offset in range(self._colspan):
if row_offset or col_offset:
self._occupied.add((self._row + row_offset, self._col + col_offset))
self._col += self._colspan
self._cell_text = None
self._rowspan = 1
self._colspan = 1
def handle_data(self, data: str) -> None:
if self._cell_text is not None:
self._cell_text.append(data)
def _positive_int(value: str | None, *, default: int) -> int:
if value is None:
return default
try:
parsed = int(value)
except ValueError:
return default
return parsed if parsed >= 1 else default
def _table_cells_from_html(value: str) -> list[TableCell]:
parser = _TableHtmlParser()
parser.feed(value)
parser.close()
return parser.cells
def _infer_benchmark_name(entries: list[ManifestEntry]) -> str:
names = sorted({str(entry.metadata.get("benchmark_name") or entry.dataset) for entry in entries})
if len(names) != 1:
raise DataValidationError(f"public benchmark manifest must contain exactly one benchmark, got: {names}")
return names[0]
def _write_public_benchmark_audit(audit: PublicBenchmarkAudit, out: Path) -> None:
out.mkdir(parents=True, exist_ok=True)
json_path = out / "public-benchmark-audit.json"
md_path = out / "public-benchmark-audit.md"
audit.summary_json_path = str(json_path)
audit.summary_md_path = str(md_path)
json_path.write_text(json.dumps(audit.to_dict(), ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
md_path.write_text(_render_public_benchmark_audit(audit), encoding="utf-8")
def _render_public_benchmark_audit(audit: PublicBenchmarkAudit) -> str:
lines = [
"# Public Benchmark Audit",
"",
f"Status: `{'pass' if audit.passed else 'fail'}`",
f"Benchmark: `{audit.benchmark}`",
f"Samples: `{audit.sample_count}`",
"",
"## Datasets",
"",
]
for dataset, count in audit.dataset_counts.items():
lines.append(f"- `{dataset}`: `{count}`")
lines.extend(["", "## Slices", ""])
for slice_name, count in audit.slice_counts.items():
lines.append(f"- `{slice_name}`: `{count}`")
if audit.issues:
lines.extend(["", "## Issues", ""])
lines.extend(f"- {issue}" for issue in audit.issues)
if audit.warnings:
lines.extend(["", "## Warnings", ""])
lines.extend(f"- {warning}" for warning in audit.warnings)
return "\n".join(lines) + "\n"
def _write_public_benchmark_prepare_report(report: PublicBenchmarkPrepareReport) -> None:
json_path = Path(report.summary_json_path)
md_path = Path(report.summary_md_path)
json_path.parent.mkdir(parents=True, exist_ok=True)
json_path.write_text(json.dumps(report.to_dict(), ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
md_path.write_text(_render_public_benchmark_prepare_report(report), encoding="utf-8")
def _render_public_benchmark_prepare_report(report: PublicBenchmarkPrepareReport) -> str:
lines = [
"# Public Benchmark Prepare Report",
"",
f"Benchmark: `{report.benchmark}`",
f"Samples: `{report.sample_count}`",
f"Manifest: `{report.manifest_path}`",
f"References: `{report.references_dir}`",
"",
"## Filters",
"",
]
for name, value in report.filters.items():
lines.append(f"- `{name}`: `{value}`")
return "\n".join(lines) + "\n"
|