Upload Generative-VQA-V2-Curated.py

Generative-VQA-V2-Curated.py

@@ -0,0 +1,95 @@
+# Generative-VQA-V2-Curated.py
+"""Custom loading script for Generative VQA V2 Curated dataset."""
+
+import datasets
+import pandas as pd
+import os
+from pathlib import Path
+
+_CITATION = """\
+@misc{devarajan_genvqa_2026,
+  author = {Devarajan},
+  title = {Generative-VQA-V2-Curated: A Balanced Dataset for Open-Ended Generative VQA},
+  year = {2026},
+  publisher = {Hugging Face},
+  howpublished = {\\url{https://huggingface.co/datasets/Deva8/Generative-VQA-V2-Curated}}
+}
+"""
+
+_DESCRIPTION = """\
+A curated, balanced, and cleaned version of the VQA v2 dataset specifically optimized for Generative Visual Question Answering.
+"""
+
+_HOMEPAGE = "https://huggingface.co/datasets/Deva8/Generative-VQA-V2-Curated"
+
+_LICENSE = "MIT"
+
+_URLS = {
+    "metadata": "main_metadata.csv",
+    "images": "gen_vqa_v2-images.zip",
+}
+
+
+class GenerativeVQAV2Curated(datasets.GeneratorBasedBuilder):
+    """Generative VQA V2 Curated dataset."""
+
+    VERSION = datasets.Version("1.0.0")
+
+    def _info(self):
+        features = datasets.Features(
+            {
+                "image_id": datasets.Value("int64"),
+                "question_id": datasets.Value("int64"),
+                "question": datasets.Value("string"),
+                "answer": datasets.Value("string"),
+                "image": datasets.Image(),
+            }
+        )
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        """Returns SplitGenerators."""
+        urls_to_download = _URLS
+        downloaded_files = dl_manager.download_and_extract(urls_to_download)
+
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "metadata_path": downloaded_files["metadata"],
+                    "images_dir": downloaded_files["images"],
+                },
+            ),
+        ]
+
+    def _generate_examples(self, metadata_path, images_dir):
+        """Yields examples."""
+        # Read metadata
+        df = pd.read_csv(metadata_path)
+
+        for idx, row in df.iterrows():
+            # Construct image path
+            # file_name format: gen_vqa_v2-images/gen_vqa_v22/images/COCO_train2014_000000429568.jpg
+            image_path = os.path.join(images_dir, row["file_name"])
+            
+            # Handle case where extraction created nested folder
+            if not os.path.exists(image_path):
+                # Try alternative path
+                alt_path = os.path.join(images_dir, "gen_vqa_v22", "images", 
+                                       os.path.basename(row["file_name"]))
+                if os.path.exists(alt_path):
+                    image_path = alt_path
+
+            yield idx, {
+                "image_id": row["image_id"],
+                "question_id": row["question_id"],
+                "question": row["question"],
+                "answer": row["answer"],
+                "image": image_path,
+            }