Spaces:
Sleeping
Sleeping
File size: 17,427 Bytes
ee838b2 c813aa1 ee838b2 | 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 | """Import de datasets OCR/HTR depuis HuggingFace Hub.
⚠ **Statut : expérimental** (phase C du chantier de refonte en 3 cercles).
L'API ``datasets`` HuggingFace évolue fréquemment et ce module n'a pas
de tests d'intégration. À utiliser à vos risques jusqu'à ce qu'un cas
d'usage institutionnel valide son comportement. Un ``UserWarning`` est
émis à l'import pour le rappeler.
Ce module fournit :
- :class:`HuggingFaceDataset` — métadonnées d'un dataset HuggingFace
- :class:`HuggingFaceImporter` — recherche et import de datasets
- :func:`search_hf_datasets` — recherche par tags dans l'API HuggingFace
- :func:`import_hf_dataset` — téléchargement d'un dataset vers un dossier local
Les datasets patrimoniaux de référence sont pré-référencés pour une découverte
rapide sans requête réseau.
Exemple
-------
importer = HuggingFaceImporter()
results = importer.search("medieval OCR", tags=["ocr"])
corpus = importer.import_dataset(results[0].dataset_id, output_dir="./corpus/")
"""
from __future__ import annotations
import json
import os
import urllib.error
import urllib.parse
import urllib.request
import warnings
from dataclasses import dataclass, field
from pathlib import Path
from typing import Optional
# Émission du warning ``experimental`` à l'import. Phase C du chantier
# de refonte — voir docstring du module ci-dessus.
warnings.warn(
"picarones.adapters.corpus.huggingface is experimental and may "
"change or be removed without notice. Use at your own risk until "
"an institutional use case validates the API.",
category=UserWarning,
stacklevel=2,
)
# ---------------------------------------------------------------------------
# Datasets de référence pré-référencés
# ---------------------------------------------------------------------------
_REFERENCE_DATASETS: list[dict] = [
{
"dataset_id": "Teklia/RIMES",
"title": "RIMES — Reconnaissance et Indexation de données Manuscrites et de fac-similEs",
"description": "Corpus de courriers manuscrits français modernes. Standard de référence pour la reconnaissance d'écriture manuscrite.",
"language": ["French"],
"tags": ["htr", "ocr", "handwritten", "french", "modern"],
"license": "cc-by-4.0",
"size_category": "1K<n<10K",
"task": "image-to-text",
"institution": "IRISA / A2iA",
"downloads": 1200,
},
{
"dataset_id": "Teklia/IAM",
"title": "IAM Handwriting Database",
"description": "Corpus de référence anglais pour la reconnaissance d'écriture manuscrite.",
"language": ["English"],
"tags": ["htr", "ocr", "handwritten", "english"],
"license": "other",
"size_category": "10K<n<100K",
"task": "image-to-text",
"institution": "University of Bern",
"downloads": 8400,
},
{
"dataset_id": "CATMuS/medieval",
"title": "CATMuS Medieval — Consistent Approaches to Transcribing ManuScripts",
"description": "Dataset multilingue de manuscrits médiévaux (latin, français, occitan, espagnol) pour l'entraînement de modèles HTR.",
"language": ["Latin", "French", "Occitan", "Spanish"],
"tags": ["htr", "medieval", "manuscripts", "latin", "french", "historical"],
"license": "cc-by-4.0",
"size_category": "100K<n<1M",
"task": "image-to-text",
"institution": "Inria / EPHE",
"downloads": 3100,
},
{
"dataset_id": "htr-united/cremma-medieval",
"title": "CREMMA Medieval",
"description": "Corpus de manuscrits médiévaux français XIIe-XVe siècles.",
"language": ["French", "Latin"],
"tags": ["htr", "medieval", "french", "manuscripts", "htr-united"],
"license": "cc-by-4.0",
"size_category": "1K<n<10K",
"task": "image-to-text",
"institution": "Inria",
"downloads": 520,
},
{
"dataset_id": "biglam/europeana_newspapers",
"title": "Europeana Newspapers",
"description": "Journaux numérisés européens du XIXe siècle (OCR + images).",
"language": ["French", "German", "Dutch", "Finnish"],
"tags": ["ocr", "newspapers", "historical", "19th-century", "europeana"],
"license": "cc0-1.0",
"size_category": "1M<n<10M",
"task": "image-to-text",
"institution": "Europeana Foundation",
"downloads": 15200,
},
{
"dataset_id": "stefanklut/esposalles",
"title": "Esposalles Dataset",
"description": "Registres de mariage catalans du XVIIe siècle pour la reconnaissance d'écriture historique.",
"language": ["Catalan", "Latin"],
"tags": ["htr", "historical", "registers", "catalan", "17th-century"],
"license": "cc-by-4.0",
"size_category": "1K<n<10K",
"task": "image-to-text",
"institution": "Universitat Autònoma de Barcelona",
"downloads": 340,
},
{
"dataset_id": "bnf-gallica/gallica-ocr",
"title": "Gallica OCR",
"description": "Extraits d'imprimés anciens numérisés depuis Gallica avec vérité terrain.",
"language": ["French", "Latin"],
"tags": ["ocr", "historical", "printed", "gallica", "french"],
"license": "etalab-2.0",
"size_category": "10K<n<100K",
"task": "image-to-text",
"institution": "Gallica",
"downloads": 2800,
},
{
"dataset_id": "Bozen-Baptism/baptism-records",
"title": "Bozen Baptism Records",
"description": "Registres de baptêmes de Bozen (Italie/Autriche) du XVIIIe siècle.",
"language": ["German", "Latin"],
"tags": ["htr", "historical", "registers", "german", "latin", "18th-century"],
"license": "cc-by-4.0",
"size_category": "1K<n<10K",
"task": "image-to-text",
"institution": "University of Innsbruck",
"downloads": 190,
},
{
"dataset_id": "read-bad/readbad",
"title": "READ-BAD — Recognition and Enrichment of Archival Documents",
"description": "Corpus multilingue de documents d'archives pour l'OCR historique (Latin, Allemand, Anglais).",
"language": ["German", "English", "Latin"],
"tags": ["ocr", "htr", "historical", "archives", "read"],
"license": "cc-by-4.0",
"size_category": "10K<n<100K",
"task": "image-to-text",
"institution": "University of Graz",
"downloads": 1050,
},
]
# ---------------------------------------------------------------------------
# Dataclass
# ---------------------------------------------------------------------------
@dataclass
class HuggingFaceDataset:
"""Métadonnées d'un dataset HuggingFace."""
dataset_id: str
title: str
description: str = ""
language: list[str] = field(default_factory=list)
tags: list[str] = field(default_factory=list)
license: str = ""
size_category: str = ""
task: str = "image-to-text"
institution: str = ""
downloads: int = 0
source: str = "reference" # "reference" | "api"
def as_dict(self) -> dict:
return {
"dataset_id": self.dataset_id,
"title": self.title,
"description": self.description,
"language": self.language,
"tags": self.tags,
"license": self.license,
"size_category": self.size_category,
"task": self.task,
"institution": self.institution,
"downloads": self.downloads,
"source": self.source,
}
@classmethod
def from_dict(cls, d: dict) -> "HuggingFaceDataset":
return cls(
dataset_id=d.get("dataset_id", d.get("id", "")),
title=d.get("title", d.get("dataset_id", "")),
description=d.get("description", ""),
language=d.get("language", []),
tags=d.get("tags", []),
license=d.get("license", ""),
size_category=d.get("size_category", d.get("cardData", {}).get("size_categories", [""])[0] if isinstance(d.get("cardData"), dict) else ""),
task=d.get("task", "image-to-text"),
institution=d.get("institution", ""),
downloads=d.get("downloads", d.get("downloadsAllTime", 0)),
source=d.get("source", "api"),
)
@property
def hf_url(self) -> str:
return f"https://huggingface.co/datasets/{self.dataset_id}"
# ---------------------------------------------------------------------------
# Importer principal
# ---------------------------------------------------------------------------
class HuggingFaceImporter:
"""Recherche et importe des datasets depuis HuggingFace Hub."""
_API_BASE = "https://huggingface.co/api"
def __init__(self, token: Optional[str] = None) -> None:
self._token = token or os.environ.get("HF_TOKEN") or os.environ.get("HUGGINGFACE_TOKEN")
def _headers(self) -> dict:
h = {"User-Agent": "picarones-hf-importer/1.0"}
if self._token:
h["Authorization"] = f"Bearer {self._token}"
return h
def search(
self,
query: str = "",
tags: Optional[list[str]] = None,
language: Optional[str] = None,
limit: int = 20,
use_reference: bool = True,
) -> list[HuggingFaceDataset]:
"""Recherche des datasets avec filtres.
Interroge d'abord les datasets de référence pré-intégrés, puis
l'API HuggingFace si disponible.
"""
results: list[HuggingFaceDataset] = []
# Datasets de référence
if use_reference:
ref_results = self._search_reference(query, tags, language)
results.extend(ref_results)
# API HuggingFace (optionnel, peut échouer silencieusement)
try:
api_results = self._search_api(query, tags, language, limit)
# Déduplique (priorité aux références)
existing_ids = {r.dataset_id for r in results}
for ds in api_results:
if ds.dataset_id not in existing_ids:
results.append(ds)
existing_ids.add(ds.dataset_id)
except Exception as exc: # noqa: BLE001 — réseau/API tierce
# Sprint A3 (B-3) : la recherche API échoue silencieusement →
# l'utilisateur ne voit que les datasets de référence et croit
# que l'API est vide. On documente l'incident.
from picarones.adapters.corpus._fallback_log import record_fallback
record_fallback(
importer="huggingface",
operation="hub_search_api",
error=exc,
extra={"query": query, "language": language, "limit": limit},
)
return results[:limit]
def _search_reference(
self,
query: str,
tags: Optional[list[str]],
language: Optional[str],
) -> list[HuggingFaceDataset]:
datasets = [HuggingFaceDataset.from_dict(d) for d in _REFERENCE_DATASETS]
datasets = [ds._replace_source("reference") for ds in datasets]
if query:
q = query.lower()
datasets = [
ds for ds in datasets
if (q in ds.title.lower()
or q in ds.description.lower()
or q in ds.dataset_id.lower()
or any(q in t.lower() for t in ds.tags)
or any(q in lg.lower() for lg in ds.language))
]
if tags:
for tag in tags:
t_lower = tag.lower()
datasets = [
ds for ds in datasets
if any(t_lower in dt.lower() for dt in ds.tags)
]
if language:
lang_lower = language.lower()
datasets = [
ds for ds in datasets
if any(lang_lower in lg.lower() for lg in ds.language)
]
return datasets
def _search_api(
self,
query: str,
tags: Optional[list[str]],
language: Optional[str],
limit: int,
) -> list[HuggingFaceDataset]:
params: dict[str, str] = {
"task_categories": "image-to-text",
"limit": str(min(limit, 50)),
"full": "False",
}
if query:
params["search"] = query
if language:
params["language"] = language
if tags:
params["tags"] = ",".join(tags)
url = f"{self._API_BASE}/datasets?" + urllib.parse.urlencode(params)
req = urllib.request.Request(url, headers=self._headers())
with urllib.request.urlopen(req, timeout=10) as resp:
data = json.loads(resp.read().decode("utf-8"))
results = []
for item in data if isinstance(data, list) else []:
ds = HuggingFaceDataset(
dataset_id=item.get("id", ""),
title=item.get("id", ""),
description=item.get("description", ""),
language=item.get("language", []),
tags=item.get("tags", []),
license=item.get("license", ""),
size_category=(
item.get("cardData", {}).get("size_categories", [""])[0]
if isinstance(item.get("cardData"), dict)
else ""
),
task="image-to-text",
downloads=item.get("downloadsAllTime", 0),
source="api",
)
if ds.dataset_id:
results.append(ds)
return results
def import_dataset(
self,
dataset_id: str,
output_dir: str | Path,
split: str = "train",
max_samples: int = 100,
show_progress: bool = True,
) -> dict:
"""Importe un dataset depuis HuggingFace vers un dossier local.
Retourne les métadonnées de l'import.
"""
output_path = Path(output_dir)
output_path.mkdir(parents=True, exist_ok=True)
meta = {
"source": "huggingface",
"dataset_id": dataset_id,
"split": split,
"max_samples": max_samples,
"imported_at": _iso_now(),
}
meta_file = output_path / "huggingface_meta.json"
meta_file.write_text(json.dumps(meta, ensure_ascii=False, indent=2), encoding="utf-8")
# Tentative d'import via datasets library si disponible
files_imported = _try_import_with_datasets_lib(
dataset_id, output_path, split, max_samples, show_progress
)
return {
"dataset_id": dataset_id,
"output_dir": str(output_path),
"files_imported": files_imported,
"metadata_file": str(meta_file),
}
def _try_import_with_datasets_lib(
dataset_id: str,
output_path: Path,
split: str,
max_samples: int,
show_progress: bool,
) -> int:
"""Essaie d'importer avec la librairie `datasets` de HuggingFace."""
try:
from datasets import load_dataset # type: ignore
ds = load_dataset(dataset_id, split=split, streaming=True)
count = 0
for i, item in enumerate(ds):
if i >= max_samples:
break
# Cherche champ image et texte
image = item.get("image") or item.get("img")
text = item.get("text") or item.get("transcription") or item.get("ground_truth", "")
if image is not None:
img_file = output_path / f"doc_{i:04d}.jpg"
try:
image.save(str(img_file))
except Exception as exc: # noqa: BLE001 — PIL/PIL-IO
# Sprint A3 (B-3) : un échec de sauvegarde d'image
# produirait un GT orphelin (texte sans image). On
# documente et on continue — le GT est tout de même
# écrit pour préserver la cohérence numérique du compteur.
from picarones.adapters.corpus._fallback_log import record_fallback
record_fallback(
importer="huggingface",
operation="image_save",
error=exc,
extra={"img_file": str(img_file), "doc_index": i},
)
gt_file = output_path / f"doc_{i:04d}.gt.txt"
gt_file.write_text(str(text), encoding="utf-8")
count += 1
return count
except (ImportError, Exception):
return 0
def _iso_now() -> str:
from datetime import datetime, timezone
return datetime.now(timezone.utc).isoformat(timespec="seconds")
# ---------------------------------------------------------------------------
# Extension de HuggingFaceDataset (helper privé)
# ---------------------------------------------------------------------------
def _patch_dataset_replace_source() -> None:
"""Ajoute un helper _replace_source à HuggingFaceDataset."""
def _replace_source(self, source: str) -> "HuggingFaceDataset":
from dataclasses import replace
return replace(self, source=source)
HuggingFaceDataset._replace_source = _replace_source
_patch_dataset_replace_source()
|