aftdb.py

import os
import json

import datasets
from PIL import Image


_DESCRIPTION = """
The Arxiv Figure Table Database (AFTdb) facilitates the linking of documentary
objects, such as figures and tables, with their captions. This enables a
comprehensive description of document-oriented images (excluding images from
cameras). For the table component, the character structure is preserved in
addition to the image of the table and its caption. This database is ideal
for multimodal processing of documentary images.
"""
_LICENSE = "apache-2.0"
_CITATION = """
@online{DeAFTdb,
  AUTHOR = {Cyrile Delestre},
  URL = {https://huggingface.co/datasets/cmarkea/aftdb},
  YEAR = {2024},
  KEYWORDS = {NLP ; Multimodal}
}
"""
_NB_TAR_FIGURE = [158, 4] # train, test
_NB_TAR_TABLE = [17, 1] # train, test


def extract_files_tar(all_path, data_dir, nb_files):
    paths = [
        os.path.join(data_dir, f"train-{ii:03d}.tar")
        for ii in range(nb_files[0])
    ]
    all_path['train'] += paths
    paths = [
        os.path.join(data_dir, f"test-{ii:03d}.tar")
        for ii in range(nb_files[1])
    ]
    all_path['test'] += paths


class AFTConfig(datasets.BuilderConfig):
    """Builder Config for AFTdb"""
 
    def __init__(self, nb_files_figure, nb_files_table, **kwargs):
        """BuilderConfig for AFTdb."""
        super().__init__(version=datasets.__version__, **kwargs)
        self.nb_files_figure = nb_files_figure
        self.nb_files_table = nb_files_table


class AFT_Dataset(datasets.GeneratorBasedBuilder):
    """Arxiv Figure Table (AFT) dataset"""

    BUILDER_CONFIGS = [
        AFTConfig(
            name="figure",
            description=(
                "Dataset containing scientific article figures associated "
                "with their caption, summary, and article title."
            ),
            data_dir="./{type}",
            nb_files_figure=_NB_TAR_FIGURE,
            nb_files_table=None
        ),
        AFTConfig(
            name="table",
            description=(
                "Dataset containing tables in JPG image format from "
                "scientific articles, along with the corresponding textual "
                "representation of the table, including its caption, summary, "
                "and article title."
            ),
            data_dir="./{type}",
            nb_files_figure=None,
            nb_files_table=_NB_TAR_TABLE
        ),
        AFTConfig(
            name="figure+table",
            description=(
                "Dataset containing figure and tables in JPG image format "
                "from scientific articles, along with the corresponding "
                "textual representation of the table, including its caption, "
                "summary, and article title."
            ),
            data_dir="./{type}",
            nb_files_figure=_NB_TAR_FIGURE,
            nb_files_table=_NB_TAR_TABLE
        )
    ]

    DEFAULT_CONFIG_NAME = "figure+table"

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    'id': datasets.Value('string'),
                    'paper_id': datasets.Value('string'),
                    'type': datasets.Value('string'),
                    'authors': datasets.Value('string'),
                    'categories': datasets.Value('string'),
                    'title': {
                        'english': datasets.Value('string'),
                        'french': datasets.Value('string')
                    },
                    'summary': {
                        'english': datasets.Value('string'),
                        'french': datasets.Value('string')
                    },
                    'caption': {
                        'english': datasets.Value('string'),
                        'french': datasets.Value('string')
                    },
                    'image': datasets.Image(),
                    'data': datasets.Value('string'),
                    'newcommands': datasets.Sequence(datasets.Value('string'))
                }
            ),
            citation=_CITATION,
            license=_LICENSE
        )

    def _split_generators(self, dl_manager: datasets.DownloadManager):
        all_path = dict(train=[], test=[])
        if self.config.nb_files_figure:
            extract_files_tar(
                all_path=all_path,
                data_dir=self.config.data_dir.format(type='figure'),
                nb_files=self.config.nb_files_figure
            )
        if self.config.nb_files_table:
            extract_files_tar(
                all_path=all_path,
                data_dir=self.config.data_dir.format(type='table'),
                nb_files=self.config.nb_files_table
            )
        if dl_manager.is_streaming:
            downloaded_files = dl_manager.download(all_path)
            downloaded_files['train'] = [
                dl_manager.iter_archive(ii) for ii in downloaded_files['train']
            ]
            downloaded_files['test'] = [
                dl_manager.iter_archive(ii) for ii in downloaded_files['test']
            ]
        else:
            downloaded_files = dl_manager.download_and_extract(all_path)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    'filepaths': downloaded_files['train'],
                    'is_streaming': dl_manager.is_streaming
                }
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "filepaths": downloaded_files['test'],
                    'is_streaming': dl_manager.is_streaming
                }
            )
        ]

    def _generate_examples(self, filepaths, is_streaming):
        if is_streaming:
            _json, _jpg = False, False
            for iter_tar in filepaths:
                for path, file_obj in iter_tar:
                    if path.endswith('.json'):
                        metadata = json.load(file_obj)
                        _json = True
                    if path.endswith('.jpg'):
                        img = Image.open(file_obj)
                        _jpg = True
                    if _json and _jpg:
                        _json, _jpg = False, False
                        yield metadata['id'], {
                            'id': metadata['id'],
                            'paper_id': metadata['paper_id'],
                            'type': metadata['type'],
                            'authors': metadata['authors'],
                            'categories': metadata['categories'],
                            'title': metadata['title'],
                            'summary': metadata['summary'],
                            'caption': metadata['caption'],
                            'image': img,
                            'data': metadata['data'],
                            'newcommands': metadata['newcommands']
                        }
        else:
            for path in filepaths:
                all_file = os.listdir(path)
                all_id_obs = sorted(
                    set(map(lambda x: x.split('.')[0], all_file))
                )
                for id_obs in all_id_obs:
                    path_metadata = os.path.join(
                        path,
                        f"{id_obs}.metadata.json"
                    )
                    path_image = os.path.join(path, f"{id_obs}.image.jpg")
                    metadata = json.load(open(path_metadata, 'r'))
                    img = Image.open(path_image)
                    yield id_obs, {
                        'id': metadata['id'],
                        'paper_id': metadata['paper_id'],
                        'type': metadata['type'],
                        'authors': metadata['authors'],
                        'categories': metadata['categories'],
                        'title': metadata['title'],
                        'summary': metadata['summary'],
                        'caption': metadata['caption'],
                        'image': img,
                        'data': metadata['data'],
                        'newcommands': metadata['newcommands']
                    }