path_vqa.py

"""PathVQA: 30000+ Questions for Medical Visual Question Answering"""

import pandas
import os

import datasets

_CITATION = """\
    @article{he2020pathvqa,
        title={PathVQA: 30000+ Questions for Medical Visual Question Answering},
        author={He, Xuehai and Zhang, Yichen and Mou, Luntian and Xing, Eric and Xie, Pengtao},
        journal={arXiv preprint arXiv:2003.10286},
        year={2020}
    }
"""

_DESCRIPTION = """\
    PathVQA is a dataset of question-answer pairs on pathology images. The questions are similar to those in the
    American Board of Pathology (ABP) test. The dataset includes both open-ended questions and binary "yes/no"
    questions. The dataset is built from two publicly-available pathology textbooks: "Textbook of Pathology" and
    "Basic Pathology", and a publicly-available digital library: "Pathology Education Informational Resource"
    (PEIR). The copyrights of images and captions belong to the publishers and authors of these two books,
    and the owners of the PEIR digital library.
"""

_HOMEPAGE = "https://github.com/UCSD-AI4H/PathVQA"

_LICENSE = "MIT"

_URLS = {
    "image_train": "data/image/train_img.tar",
    "image_val": "data/image/val_img.tar",
    "image_test": "data/image/test_img.tar",
    "text_train": "data/text/train_qa.jsonl",
    "text_val": "data/text/val_qa.jsonl",
    "text_test": "data/text/test_qa.jsonl",
}

class PathVQA(datasets.GeneratorBasedBuilder):
    
    """
    PathVQA: 30000+ Questions for Medical Visual Question Answering.
    
    The data was obtained from the updated Google Drive link shared by the authors in their GitHub repository
    on Feb 15, 2023, see https://github.com/UCSD-AI4H/PathVQA/commit/117e7f4ef88a0e65b0e7f37b98a73d6237a3ceab.
    
    This version of the dataset contains a total of 5,004 images and 32,795 question-answer pairs. Of the
    5,004 images, 4,289 images are referenced by a question-answer pair, while 715 images are not used.
    Furthermore, there are several duplicates, i.e. there are some image-question-answer triplets which occur
    more than once in the same split (train, val, test). After dropping the duplicate image-question-answer
    triplets, the dataset contains 32,632 question-answer pairs on 4,289 images.
    """

    VERSION = datasets.Version("0.1.0")
    
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(name="full", version=VERSION, description="Original dataset."),
        datasets.BuilderConfig(name="de-duped", version=VERSION, description="De-duplicated dataset."),
    ]

    DEFAULT_CONFIG_NAME = "de-duped"
    
    def _info(self):
        
        features = datasets.Features(
            {
                "image": datasets.Image(),
                "question": datasets.Value("string"),
                "answer": datasets.Value("string")
            }
        )
        
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        
        # images
        image_train_dir = dl_manager.download_and_extract(_URLS["image_train"])
        image_val_dir = dl_manager.download_and_extract(_URLS["image_val"])
        image_test_dir = dl_manager.download_and_extract(_URLS["image_test"])

        # question-answer pairs
        text_train_dir = dl_manager.download(_URLS["text_train"])
        text_val_dir = dl_manager.download(_URLS["text_val"])
        text_test_dir = dl_manager.download(_URLS["text_test"])

        return [
            
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "image_filepath": os.path.join(image_train_dir),
                    "text_filepath": os.path.join(text_train_dir),
                    "split": "train",
                },
            ),
            
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "image_filepath": os.path.join(image_val_dir),
                    "text_filepath": os.path.join(text_val_dir),
                    "split": "val",
                },
            ),
            
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "image_filepath": os.path.join(image_test_dir),
                    "text_filepath": os.path.join(text_test_dir),
                    "split": "test"
                },
            ),
        ]

    def _generate_examples(self, image_filepath, text_filepath, split):
        
        df = pandas.read_json(text_filepath, orient='records', lines=True)
        if self.config.name == "de-duped":
            df = df.drop_duplicates(ignore_index=True)
        
        for key, row in df.iterrows():
            yield key, {
                "image": os.path.join(image_filepath, row['image']),
                "question": row["question"],
                "answer": row["answer"]
            }