kmnist.py

import struct

import numpy as np

import datasets
from datasets.tasks import ImageClassification

_CITATION = R"""
@article{DBLP:journals/corr/abs-1812-01718,
  author       = {Tarin Clanuwat and
                  Mikel Bober{-}Irizar and
                  Asanobu Kitamoto and
                  Alex Lamb and
                  Kazuaki Yamamoto and
                  David Ha},
  title        = {Deep Learning for Classical Japanese Literature},
  journal      = {CoRR},
  volume       = {abs/1812.01718},
  year         = {2018},
  url          = {http://arxiv.org/abs/1812.01718},
  eprinttype    = {arXiv},
  eprint       = {1812.01718},
  timestamp    = {Thu, 14 Oct 2021 09:15:14 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/abs-1812-01718.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}
"""

_URL = "./raw/"
_URLS = {
    "train_images": "train-images-idx3-ubyte.gz",
    "train_labels": "train-labels-idx1-ubyte.gz",
    "test_images": "t10k-images-idx3-ubyte.gz",
    "test_labels": "t10k-labels-idx1-ubyte.gz",
}


class KMNIST(datasets.GeneratorBasedBuilder):

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="kmnist",
            version=datasets.Version("1.0.0"),
        )
    ]

    def _info(self):
        return datasets.DatasetInfo(
            features=datasets.Features(
                {
                    "image": datasets.Image(),
                    "label": datasets.features.ClassLabel(
                        names=[
                            "お",
                            "き",
                            "す",
                            "つ",
                            "な",
                            "は",
                            "ま",
                            "や",
                            "れ",
                            "を",
                        ]
                    ),
                }
            ),
            supervised_keys=("image", "label"),
            homepage="https://github.com/rois-codh/kmnist",
            citation=_CITATION,
            task_templates=[
                ImageClassification(
                    image_column="image",
                    label_column="label",
                )
            ],
        )

    def _split_generators(self, dl_manager):
        urls_to_download = {key: _URL + fname for key, fname in _URLS.items()}
        downloaded_files = dl_manager.download_and_extract(urls_to_download)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "filepath": (
                        downloaded_files["train_images"],
                        downloaded_files["train_labels"],
                    ),
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "filepath": (
                        downloaded_files["test_images"],
                        downloaded_files["test_labels"],
                    ),
                    "split": "test",
                },
            ),
        ]

    def _generate_examples(self, filepath, split):
        """This function returns the examples in the raw form."""
        # Images
        with open(filepath[0], "rb") as f:
            # First 16 bytes contain some metadata
            _ = f.read(4)
            size = struct.unpack(">I", f.read(4))[0]
            _ = f.read(8)
            images = np.frombuffer(f.read(), dtype=np.uint8).reshape(size, 28, 28)

        # Labels
        with open(filepath[1], "rb") as f:
            # First 8 bytes contain some metadata
            _ = f.read(8)
            labels = np.frombuffer(f.read(), dtype=np.uint8)

        for idx in range(size):
            yield idx, {"image": images[idx], "label": str(labels[idx])}