feat: add loading script

patfig.py

@@ -0,0 +1,69 @@
+import datasets
+from datasets import load_dataset, Dataset, Value, Sequence, Features, DatasetInfo, GeneratorBasedBuilder, Image
+
+from pathlib import Path
+import os
+import pandas as pd
+
+_DESCRIPTION = """\ The PatFig Dataset is a curated collection of over 18,000 patent images from more than 7,
+000 European patent applications, spanning the year 2020. It aims to provide a comprehensive resource for research 
+and applications in image captioning, abstract reasoning, patent analysis, and automated documentprocessing. The 
+overarching goal of this dataset is to advance the research in visually situated language understanding towards more 
+hollistic consumption of the visual and textual data.
+"""
+
+_URL = "https://huggingface.co/datasets/lcolonn/patfig/resolve/main/"
+_URLS = {
+    "train_images": "train/train_images.tar.gz",
+    "test_images": "test/test_images.tar.gz",
+    "annotations_train": "train/annotations_train.parquet",
+    "annotations_test": "test/annotations_test.parquet",
+}
+
+
+class PatFig(GeneratorBasedBuilder):
+    """DatasetBuilder for patfig dataset."""
+
+    def _info(self):
+        return DatasetInfo(
+            description=_DESCRIPTION,
+            features=Features({
+                "image": Image(),
+                "image_name": Value("string"),
+                "pub_number": Value("string"),
+                "title": Value("string"),
+                "figs_norm": Sequence(feature=Value("string"), length=-1),
+                "short_description": Sequence(feature=Value("string"), length=-1),
+                "long_description": Sequence(feature=Value("string"), length=-1),
+                "short_description_token_count": Value("int64"),
+                "long_description_token_count": Value("int64"),
+                "draft_class": Value("string"),
+                "cpc_class": Value("string"),
+                "relevant_terms": [{'element_identifier': Value("string"), "terms": Sequence(feature=Value("string"), length=-1)}],
+                "associated_claims": Value("string"),
+                "compound": Value("bool"),
+                "references": Sequence(feature=Value(dtype='string'), length=-1),
+            }),
+        )
+
+    def _split_generators(self, dl_manager: datasets.DownloadManager):
+        urls_to_download = {key: _URL + fname for key, fname in _URLS.items()}
+        downloaded_files = dl_manager.download_and_extract(urls_to_download)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN, gen_kwargs={"images_dir": downloaded_files["train_images"], "annotations_dir": downloaded_files["annotations_train"]}
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.TEST, gen_kwargs={"images_dir": f'{downloaded_files["test_images"]}', "annotations_dir": downloaded_files["annotations_test"]}
+            ),
+        ]
+
+    def _generate_examples(self, images_dir: str, annotations_dir: str):
+        df = pd.read_csv(annotations_dir)
+
+        for idx, row in df.iterrows():
+            image_path = os.path.join(images_dir, row["pub_number"], row["image_name"])
+            yield idx, {
+                "image": image_path,
+                **row.to_dict(),
+            }