rvl_cdip_n_mp.py

# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""RVL-CDIP (Ryerson Vision Lab Complex Document Information Processing) dataset"""


import os
import datasets
from pathlib import Path
from tqdm import tqdm
import pdf2image

datasets.logging.set_verbosity_info()
logger = datasets.logging.get_logger(__name__)

_MODE = "binary"

_CITATION = """\
@inproceedings{larson2022evaluating,
	title={Evaluating Out-of-Distribution Performance on Document Image Classifiers},
	author={Larson, Stefan and Lim, Gordon and Ai, Yutong and Kuang, David and Leach, Kevin},
	booktitle={Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track},
	year={2022}
}
    
@inproceedings{bdpc,
    title = {Beyond Document Page Classification},
    author = {Anonymous},
    booktitle = {Under Review},
    year = {2023}
}
"""

_DESCRIPTION = """\
The RVL-CDIP-N (Ryerson Vision Lab Complex Document Information Processing) dataset consists of newly gathered documents in 16 classes 
There are 998 documents for testing purposes. There were 3 documents from the original dataset that could not be retrieved based on the metadata. 
"""


_HOMEPAGE = "https://www.cs.cmu.edu/~aharley/rvl-cdip/"
_LICENSE = "https://www.industrydocuments.ucsf.edu/help/copyright/"


SOURCE = "jordyvl/rvl_cdip_mp"
_URL = f"https://huggingface.co/datasets/{SOURCE}/resolve/main/data.gz"
_BACKOFF_folder = "/mnt/lerna/data/RVL-CDIP-NO/RVL-CDIP-N_pdf/data"

_CLASSES = [
    "letter",
    "form",
    "email",
    "handwritten",
    "advertisement",
    "scientific report",
    "scientific publication",
    "specification",
    "file folder",
    "news article",
    "budget",
    "invoice",
    "presentation",
    "questionnaire",
    "resume",
    "memo",
]


def batched_conversion(pdf_file):
    info = pdf2image.pdfinfo_from_path(pdf_file, userpw=None, poppler_path=None)
    maxPages = info["Pages"]

    logger.info(f"{pdf_file} has {str(maxPages)} pages")

    images = []

    for page in range(1, maxPages + 1, 10):
        images.extend(
            pdf2image.convert_from_path(pdf_file, dpi=200, first_page=page, last_page=min(page + 10 - 1, maxPages))
        )
    return images


def open_pdf_binary(pdf_file):
    with open(pdf_file, "rb") as f:
        return f.read()


class RvlCdipNMp(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("1.0.0")
    DEFAULT_CONFIG_NAME = "default"

    def _info(self):
        if isinstance(self.config.data_dir, str):
            folder = self.config.data_dir  # contains the folder structure at someone local disk
        else:
            if not os.path.exists(_BACKOFF_folder):
                raise ValueError("No data folder found. Please set data_dir or data_files.")
            folder = _BACKOFF_folder  # my local path, others should set data_dir or data_files
        self.config.data_dir = folder

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "id": datasets.Value("string"),
                    "file": datasets.Value("binary"),
                    "labels": datasets.features.ClassLabel(names=_CLASSES),
                }
            ),
            homepage=_HOMEPAGE,
            citation=_CITATION,
            license=_LICENSE,
            task_templates=None,
        )

    def _split_generators(self, dl_manager):
        if self.config.data_dir.endswith(".tar.gz"):
            archive_path = dl_manager.download(self.config.data_dir)
            data_files = dl_manager.iter_archive(archive_path)
        else:
            data_files = self.config.data_dir

        return [datasets.SplitGenerator(name="test", gen_kwargs={"archive_path": data_files})]

    def _generate_examples(self, archive_path):
        labels = self.info.features["labels"]

        extensions = {".pdf", ".PDF"}

        for i, path in tqdm(enumerate(Path(archive_path).glob("**/*")), desc=f"{archive_path}"):
            if path.suffix in extensions:
                try:
                    if _MODE == "binary":
                        images = open_pdf_binary(path)
                        # batched_conversion(path)
                    else:
                        images = path

                    a = dict(
                        id=path.name,
                        file=images,
                        labels=labels.encode_example(path.parent.name.lower()),
                    )
                    yield path.name, a

                except Exception as e:
                    logger.warning(f"{e} failed to parse {i}")