| import json | |
| import os | |
| from datasets import DatasetInfo, GeneratorBasedBuilder, SplitGenerator, Split, Value, Features, Image | |
| class AugeoBench(GeneratorBasedBuilder): | |
| def _info(self): | |
| return DatasetInfo( | |
| description="AugeoBench: Multimodal QA dataset with Japanese problem-solving questions and diagram images.", | |
| features=Features({ | |
| "id": Value("string"), | |
| "url": Value("string"), | |
| "question_context_ja": Value("string"), | |
| "question_text_ja": Value("string"), | |
| "question_information_ja": Value("string"), | |
| "answer_exact": Value("string"), | |
| "answer_text_ja": Value("string"), | |
| "question_image": Image(), | |
| "answer_image": Image(), | |
| "Genre": Value("string"), | |
| "Remarks": Value("string"), | |
| }), | |
| supervised_keys=None, | |
| ) | |
| def _split_generators(self, dl_manager): | |
| data_files = { | |
| "annotation": "annotation_clean.json", | |
| "images": "images.zip", | |
| } | |
| downloaded_files = dl_manager.download_and_extract(data_files) | |
| return [ | |
| SplitGenerator( | |
| name=Split.TRAIN, | |
| gen_kwargs={ | |
| "annotation_path": downloaded_files["annotation"], | |
| "images_dir": downloaded_files["images"], | |
| }, | |
| ) | |
| ] | |
| def _generate_examples(self, annotation_path, images_dir): | |
| with open(annotation_path, encoding="utf-8") as f: | |
| data = json.load(f) | |
| for idx, item in enumerate(data): | |
| q_img = os.path.join(images_dir, "images", item["question_image_path"]) if item.get("question_image_path") else None | |
| a_img = os.path.join(images_dir, "images", item["answer_image_path"]) if item.get("answer_image_path") else None | |
| yield idx, { | |
| "id": item["id"], | |
| "url": item["url"], | |
| "question_context_ja": item["question_context_ja"], | |
| "question_text_ja": item["question_text_ja"], | |
| "question_information_ja": item["question_information_ja"], | |
| "answer_exact": item["answer_exact"], | |
| "answer_text_ja": item["answer_text_ja"], | |
| "question_image": q_img, | |
| "answer_image": a_img, | |
| "Genre": item["Genre"], | |
| "Remarks": item["Remarks"], | |
| } | |