Datasets:
Delete forbin_dataset.py
Browse files- forbin_dataset.py +0 -176
forbin_dataset.py
DELETED
|
@@ -1,176 +0,0 @@
|
|
| 1 |
-
from collections import defaultdict
|
| 2 |
-
import json
|
| 3 |
-
import os
|
| 4 |
-
import datasets
|
| 5 |
-
from datasets import (
|
| 6 |
-
Value,
|
| 7 |
-
Image,
|
| 8 |
-
Sequence,
|
| 9 |
-
ClassLabel,
|
| 10 |
-
)
|
| 11 |
-
from typing import Dict, Any
|
| 12 |
-
import tarfile
|
| 13 |
-
from io import BytesIO
|
| 14 |
-
|
| 15 |
-
_DESCRIPTION = """
|
| 16 |
-
Forbin Dataset — a curated collection of heritage photographs and documents, combining digitized facsimiles, historical context, and fine-grained annotations.
|
| 17 |
-
|
| 18 |
-
This collection brings together images from the archives of the French explorer and writer Victor Forbin (1868–1947), enhanced with semantic structures following the COCO Segmentation format for scientific reuse.
|
| 19 |
-
"""
|
| 20 |
-
_HOMEPAGE = "https://huggingface.co/mchelali/forbin_dataset"
|
| 21 |
-
_LICENSE = "CC-BY-NC-4.0" # Licence de votre choix
|
| 22 |
-
_ANNOTATIONS_JSON = "forbin_annotated.json"
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
def _load_image_from_tar(
|
| 26 |
-
dl_manager: datasets.DownloadManager,
|
| 27 |
-
repo_url: str,
|
| 28 |
-
tar_filename: str,
|
| 29 |
-
image_key_in_tar: str,
|
| 30 |
-
) -> BytesIO:
|
| 31 |
-
"""
|
| 32 |
-
Fonction CRITIQUE : Ouvre un fichier .tar distant en streaming et extrait un fichier spécifique par sa clé.
|
| 33 |
-
"""
|
| 34 |
-
tar_url = f"{repo_url}/{tar_filename}"
|
| 35 |
-
|
| 36 |
-
# 1. Ouverture du fichier .tar en streaming via l'outil de téléchargement de datasets
|
| 37 |
-
# Cela évite de télécharger le fichier en entier.
|
| 38 |
-
with dl_manager.open(tar_url) as f:
|
| 39 |
-
# 2. Utilisation de tarfile pour lire le flux. Le mode 'r|*' est pour la lecture par flux (non-seekable).
|
| 40 |
-
with tarfile.open(fileobj=f, mode="r|*") as tar:
|
| 41 |
-
try:
|
| 42 |
-
# 3. Recherche du membre (l'image) par son chemin exact à l'intérieur du .tar
|
| 43 |
-
member = tar.getmember(image_key_in_tar)
|
| 44 |
-
except KeyError:
|
| 45 |
-
raise FileNotFoundError(
|
| 46 |
-
f"Image '{image_key_in_tar}' non trouvée dans {tar_filename}."
|
| 47 |
-
)
|
| 48 |
-
|
| 49 |
-
# 4. Extraction du contenu binaire de l'image
|
| 50 |
-
image_file = tar.extractfile(member)
|
| 51 |
-
if image_file is None:
|
| 52 |
-
raise Exception(
|
| 53 |
-
f"Impossible d'extraire le fichier pour {image_key_in_tar}."
|
| 54 |
-
)
|
| 55 |
-
|
| 56 |
-
# 5. Lecture du contenu dans un objet BytesIO pour la feature Image()
|
| 57 |
-
return BytesIO(image_file.read())
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
class ForbinDataset(datasets.GeneratorBasedBuilder):
|
| 61 |
-
|
| 62 |
-
VERSION = datasets.Version("1.0.0")
|
| 63 |
-
|
| 64 |
-
# Définir le fichier de données (le manifeste Parquet)
|
| 65 |
-
DEFAULT_CONFIG_NAME = "default"
|
| 66 |
-
BUILDER_CONFIGS = [
|
| 67 |
-
datasets.BuilderConfig(
|
| 68 |
-
name=DEFAULT_CONFIG_NAME,
|
| 69 |
-
version=VERSION,
|
| 70 |
-
description=_DESCRIPTION,
|
| 71 |
-
)
|
| 72 |
-
]
|
| 73 |
-
|
| 74 |
-
def _info(self) -> datasets.DatasetInfo:
|
| 75 |
-
# Définition des catégories
|
| 76 |
-
categories = ["legende", "texte", "image", "tampon", "etc"]
|
| 77 |
-
|
| 78 |
-
features = datasets.Features(
|
| 79 |
-
{
|
| 80 |
-
"image_id": Value("int32"),
|
| 81 |
-
# Utiliser Image() avec les URL générées (pipe:curl...)
|
| 82 |
-
"recto_image": Image(),
|
| 83 |
-
"verso_image": Image(),
|
| 84 |
-
# "tar_file": Value("string"),
|
| 85 |
-
# "dpi": Value("float32"),
|
| 86 |
-
"Carton": Value("string"),
|
| 87 |
-
"Pays": Value("string"),
|
| 88 |
-
# ...
|
| 89 |
-
"annotations": Sequence(
|
| 90 |
-
{
|
| 91 |
-
# ... Définition des champs d'annotations
|
| 92 |
-
"id": Value("int32"),
|
| 93 |
-
"category_id": ClassLabel(names=categories),
|
| 94 |
-
"bbox": Sequence(Value("float32")),
|
| 95 |
-
"text": Value("string"),
|
| 96 |
-
# ...
|
| 97 |
-
}
|
| 98 |
-
),
|
| 99 |
-
}
|
| 100 |
-
)
|
| 101 |
-
return datasets.DatasetInfo(description=_DESCRIPTION, features=features)
|
| 102 |
-
|
| 103 |
-
def _split_generators(
|
| 104 |
-
self, dl_manager: datasets.DownloadManager
|
| 105 |
-
) -> list[datasets.SplitGenerator]:
|
| 106 |
-
|
| 107 |
-
urls = {"annotations": os.path.join(dl_manager.manual_dir, _ANNOTATIONS_JSON)}
|
| 108 |
-
downloaded_files = dl_manager.download(urls)
|
| 109 |
-
|
| 110 |
-
return [
|
| 111 |
-
datasets.SplitGenerator(
|
| 112 |
-
name=datasets.Split.TRAIN,
|
| 113 |
-
gen_kwargs={"annotations_filepath": downloaded_files["annotations"]},
|
| 114 |
-
)
|
| 115 |
-
]
|
| 116 |
-
|
| 117 |
-
def _generate_examples(self, annotations_filepath: str):
|
| 118 |
-
"""Lit le manifeste Parquet et génère les exemples."""
|
| 119 |
-
|
| 120 |
-
# Lire le fichier Parquet
|
| 121 |
-
repo_id = "mchelali/forbin_dataset"
|
| 122 |
-
repo_url = f"https://huggingface.co/datasets/{repo_id}/resolve/main/data"
|
| 123 |
-
|
| 124 |
-
with open(annotations_filepath, "r", encoding="utf-8") as f:
|
| 125 |
-
annotations = json.load(f)
|
| 126 |
-
|
| 127 |
-
for _id, record in enumerate(annotations):
|
| 128 |
-
|
| 129 |
-
# Conversion de la chaîne JSON des annotations en objet Python
|
| 130 |
-
# --- STRUCTURE DE VOTRE ENREGISTREMENT JSON (À ADAPTER) ---
|
| 131 |
-
image_id = record.get("image_id", str(_id))
|
| 132 |
-
annotations_data = record.get(
|
| 133 |
-
"annotations", {}
|
| 134 |
-
) # Votre champ d'annotations
|
| 135 |
-
recto_path_in_tar = record["recto_path"] # Ex: SHDGR.../0001.jpg
|
| 136 |
-
verso_path_in_tar = record["verso_path"] # Ex: SHDGR.../0002.jpg
|
| 137 |
-
# -----------------------------------------------------------
|
| 138 |
-
|
| 139 |
-
# 2. Déduction du nom de l'archive .tar (le dossier parent ou la clé principale)
|
| 140 |
-
# Cette logique suppose que le nom du dossier parent de l'image est aussi la base du nom du .tar
|
| 141 |
-
tar_name = (
|
| 142 |
-
recto_path_in_tar.split(os.path.sep)[2] + ".tar"
|
| 143 |
-
) # Ex: SHDGR..._2.tar
|
| 144 |
-
|
| 145 |
-
try:
|
| 146 |
-
# 3. Chargement de l'image Recto
|
| 147 |
-
recto_image_bytes = _load_image_from_tar(
|
| 148 |
-
self.dl_manager, repo_url, tar_name, recto_path_in_tar
|
| 149 |
-
)
|
| 150 |
-
|
| 151 |
-
# 4. Chargement de l'image Verso
|
| 152 |
-
verso_image_bytes = _load_image_from_tar(
|
| 153 |
-
self.dl_manager, repo_url, tar_name, verso_path_in_tar
|
| 154 |
-
)
|
| 155 |
-
|
| 156 |
-
# 5. Génération de l'exemple
|
| 157 |
-
yield _id, {
|
| 158 |
-
"id": image_id,
|
| 159 |
-
# L'Image feature de datasets utilise le dict {"path": ..., "bytes": ...} pour charger l'image.
|
| 160 |
-
"recto_image": {
|
| 161 |
-
"path": recto_path_in_tar,
|
| 162 |
-
"bytes": recto_image_bytes,
|
| 163 |
-
},
|
| 164 |
-
"verso_image": {
|
| 165 |
-
"path": verso_path_in_tar,
|
| 166 |
-
"bytes": verso_image_bytes,
|
| 167 |
-
},
|
| 168 |
-
"annotations": json.dumps(annotations_data),
|
| 169 |
-
}
|
| 170 |
-
|
| 171 |
-
except Exception as e:
|
| 172 |
-
# Gérer les erreurs de fichiers manquants ou de lecture
|
| 173 |
-
print(f"ERREUR (Exemple {_id}): {e}")
|
| 174 |
-
continue
|
| 175 |
-
|
| 176 |
-
_id += 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|