openvivqa.py

# coding=utf-8
import json
import os
from pathlib import Path
from typing import Dict, List, Tuple

import datasets

from seacrowd.utils import schemas
from seacrowd.utils.configs import SEACrowdConfig
from seacrowd.utils.constants import Licenses, Tasks

_CITATION = """\
@inproceedings{tran2021vivqa,
  title={ViVQA: Vietnamese visual question answering},
  author={Tran, Khanh Quoc and Nguyen, An Trong and Le, An Tran-Hoai and Van Nguyen, Kiet},
  booktitle={Proceedings of the 35th Pacific Asia Conference on Language, Information and Computation},
  pages={683--691},
  year={2021}
}
"""
_DATASETNAME = "openvivqa"
_DESCRIPTION = """\
OpenViVQA (Open-domain Vietnamese Visual Question Answering) is a dataset for VQA (Visual Question Answering) with
open-ended answers in Vietnamese. It consisted of 11199 images associated with 37914 question-answer pairs (QAs).
Images in the OpenViVQA dataset are captured in Vietnam and question-answer pairs are created manually by Vietnamese
crowd workers.
"""
_HOMEPAGE = "https://huggingface.co/datasets/uitnlp/OpenViVQA-dataset"
_LANGUAGES = ["vie"]
_LICENSE = Licenses.MIT.value
_LOCAL = False
_HF_URL = "https://huggingface.co/datasets/uitnlp/OpenViVQA-dataset"
_URLS = {
    "dataset": {
        "train": "https://huggingface.co/datasets/uitnlp/OpenViVQA-dataset/raw/main/vlsp2023_train_data.json",
        "test": "https://huggingface.co/datasets/uitnlp/OpenViVQA-dataset/raw/main/vlsp2023_test_data.json",
        "dev": "https://huggingface.co/datasets/uitnlp/OpenViVQA-dataset/raw/main/vlsp2023_dev_data.json",
    },
    "images": {
        "train": "https://huggingface.co/datasets/uitnlp/OpenViVQA-dataset/resolve/main/train-images.zip?download=true",
        "test": "https://huggingface.co/datasets/uitnlp/OpenViVQA-dataset/resolve/main/test-images.zip?download=true",
        "dev": "https://huggingface.co/datasets/uitnlp/OpenViVQA-dataset/resolve/main/dev-images.zip?download=true",
    },
}
_SUPPORTED_TASKS = [Tasks.VISUAL_QUESTION_ANSWERING]
_SOURCE_VERSION = "1.0.0"
_SEACROWD_VERSION = "2024.06.20"


class OpenViVQADataset(datasets.GeneratorBasedBuilder):

    SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
    SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)

    BUILDER_CONFIGS = [
        SEACrowdConfig(
            name=f"{_DATASETNAME}_source",
            version=SOURCE_VERSION,
            description=f"{_DATASETNAME} source schema",
            schema="source",
            subset_id=f"{_DATASETNAME}",
        ),
        SEACrowdConfig(
            name=f"{_DATASETNAME}_seacrowd_imqa",
            version=SEACROWD_VERSION,
            description=f"{_DATASETNAME} SEACrowd schema",
            schema="seacrowd_imqa",
            subset_id=f"{_DATASETNAME}",
        ),
    ]

    DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source"

    def _info(self) -> datasets.DatasetInfo:

        if self.config.schema == "source":
            features = datasets.Features({"img_path": datasets.Value("string"),
                                          "question": datasets.Value("string"),
                                          "answer": datasets.Value("string"),
                                          "id": datasets.Value("string")})
        elif self.config.schema == "seacrowd_imqa":
            features = schemas.imqa_features
            # features["meta"] = {"image_path": datasets.Value("string")}
        else:
            raise ValueError(f"No schema matched for {self.config.schema}")

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
        """Returns SplitGenerators."""
        data_dir = dl_manager.download_and_extract(_URLS["dataset"])
        image_dir = dl_manager.download_and_extract(_URLS["images"])

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "filepath": data_dir["train"],
                    "imagepath": os.path.join(image_dir["train"], "training-images"),
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "filepath": data_dir["test"],
                    "imagepath": os.path.join(image_dir["test"], "test-images"),
                    "split": "test",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "filepath": data_dir["dev"],
                    "imagepath": os.path.join(image_dir["dev"], "dev-images"),
                    "split": "validation",
                },
            ),
        ]

    def _generate_examples(self, filepath: Path, imagepath: Path, split: str) -> Tuple[int, Dict]:
        """Yields examples as (key, example) tuples."""

        raw_examples = json.load(open(filepath, "r"))
        images = raw_examples["images"]
        data_annotations = raw_examples["annotations"]
        for sample_id, q_key in enumerate(list(data_annotations.keys())):
            quest_id = q_key
            sample = data_annotations[q_key]
            sample_img_id = sample["image_id"]
            sample_img_name = images[str(sample_img_id)]
            sample_img_path = os.path.join(imagepath, sample_img_name)
            sample_question = sample["question"]
            sample_answer = sample["answer"]
            if self.config.schema == "source":
                example = {
                    "img_path": sample_img_path,
                    "question": sample_question,
                    "answer": sample_answer,
                    "id": quest_id,
                }
            elif self.config.schema == "seacrowd_imqa":
                example = {
                    "id": q_key,
                    "question_id": q_key,
                    "document_id": q_key,
                    "questions": [sample_question],
                    "type": None,
                    "choices": None,
                    "context": sample_img_id,
                    "answer": [sample_answer],
                    "image_paths": [sample_img_path],
                    "meta": {},
                }
            yield sample_id, example