cartoonset.py

"""Cartoonset-10k Data Set"""


from io import BytesIO
from typing import Optional

import tarfile
import pandas as pd


import datasets


_CITATION = r"""
@article{DBLP:journals/corr/abs-1711-05139,
  author    = {Amelie Royer and
               Konstantinos Bousmalis and
               Stephan Gouws and
               Fred Bertsch and
               Inbar Mosseri and
               Forrester Cole and
               Kevin Murphy},
  title     = {{XGAN:} Unsupervised Image-to-Image Translation for many-to-many Mappings},
  journal   = {CoRR},
  volume    = {abs/1711.05139},
  year      = {2017},
  url       = {http://arxiv.org/abs/1711.05139},
  eprinttype = {arXiv},
  eprint    = {1711.05139},
  timestamp = {Mon, 13 Aug 2018 16:47:38 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-1711-05139.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}
"""

_DESCRIPTION = """\
Cartoon Set is a collection of random, 2D cartoon avatar images. The cartoons vary in 10 artwork 
categories, 4 color categories, and 4 proportion categories, with a total of ~1013 possible 
combinations. We provide sets of 10k and 100k randomly chosen cartoons and labeled attributes. 
"""

_DATA_URLS = {
    "10k": "https://huggingface.co/datasets/cgarciae/cartoonset/resolve/1.0.0/data/cartoonset10k.tgz",
    "100k": "https://huggingface.co/datasets/cgarciae/cartoonset/resolve/1.0.0/data/cartoonset100k.tgz",
}


class Cartoonset(datasets.GeneratorBasedBuilder):
    """Cartoonset-10k Data Set"""

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="10k",
            version=datasets.Version("1.0.0", ""),
            description="Loads the Cartoonset-10k Data Set (images only).",
        ),
        datasets.BuilderConfig(
            name="10k+features",
            version=datasets.Version("1.0.0", ""),
            description="Loads the Cartoonset-10k Data Set (images and attributes).",
        ),
        datasets.BuilderConfig(
            name="100k",
            version=datasets.Version("1.0.0", ""),
            description="Loads the Cartoonset-100k Data Set (images only).",
        ),
        datasets.BuilderConfig(
            name="100k+features",
            version=datasets.Version("1.0.0", ""),
            description="Loads the Cartoonset-100k Data Set (images and attributes).",
        ),
    ]

    DEFAULT_CONFIG_NAME = "10k"

    def _info(self):
        features = {"img_bytes": datasets.Value("binary")}

        if self.config.name.endswith("+features"):
            features.update(
                {
                    "eye_angle": datasets.Value("int32"),
                    "eye_angle_num_categories": datasets.Value("int32"),
                    "eye_lashes": datasets.Value("int32"),
                    "eye_lashes_num_categories": datasets.Value("int32"),
                    "eye_lid": datasets.Value("int32"),
                    "eye_lid_num_categories": datasets.Value("int32"),
                    "chin_length": datasets.Value("int32"),
                    "chin_length_num_categories": datasets.Value("int32"),
                    "eyebrow_weight": datasets.Value("int32"),
                    "eyebrow_weight_num_categories": datasets.Value("int32"),
                    "eyebrow_shape": datasets.Value("int32"),
                    "eyebrow_shape_num_categories": datasets.Value("int32"),
                    "eyebrow_thickness": datasets.Value("int32"),
                    "eyebrow_thickness_num_categories": datasets.Value("int32"),
                    "face_shape": datasets.Value("int32"),
                    "face_shape_num_categories": datasets.Value("int32"),
                    "facial_hair": datasets.Value("int32"),
                    "facial_hair_num_categories": datasets.Value("int32"),
                    "hair": datasets.Value("int32"),
                    "hair_num_categories": datasets.Value("int32"),
                    "eye_color": datasets.Value("int32"),
                    "eye_color_num_categories": datasets.Value("int32"),
                    "face_color": datasets.Value("int32"),
                    "face_color_num_categories": datasets.Value("int32"),
                    "hair_color": datasets.Value("int32"),
                    "hair_color_num_categories": datasets.Value("int32"),
                    "glasses": datasets.Value("int32"),
                    "glasses_num_categories": datasets.Value("int32"),
                    "glasses_color": datasets.Value("int32"),
                    "glasses_color_num_categories": datasets.Value("int32"),
                    "eye_slant": datasets.Value("int32"),
                    "eye_slant_num_categories": datasets.Value("int32"),
                    "eyebrow_width": datasets.Value("int32"),
                    "eyebrow_width_num_categories": datasets.Value("int32"),
                    "eye_eyebrow_distance": datasets.Value("int32"),
                    "eye_eyebrow_distance_num_categories": datasets.Value("int32"),
                }
            )

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(features),
            supervised_keys=("img_bytes",),
            homepage="https://www.cs.toronto.edu/~kriz/cifar.html",
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager: datasets.DownloadManager):

        url = _DATA_URLS[self.config.name.replace("+features", "")]
        archive = dl_manager.download(url)

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "files": dl_manager.iter_archive(archive),
                    "split": "train",
                },
            ),
        ]

    def _generate_examples(self, files, split):
        """This function returns the examples in the raw (text) form."""

        if self.config.name.endswith("+features"):
            return self._generate_examples_with_features(files, split)
        else:
            return self._generate_examples_without_features(files, split)

    def _generate_examples_without_features(self, files, split):
        path: str
        file_obj: tarfile.ExFileObject
        root: str
        for path, file_obj in files:
            root = path[:-4]

            if path.endswith(".png"):
                image = file_obj.read()

                yield root, {"img_bytes": image}

    def _generate_examples_with_features(self, files, split):
        path: str
        file_obj: tarfile.ExFileObject
        outputs = {}
        root: Optional[str] = None
        for path, file_obj in files:
            root = path[:-4]

            if root not in outputs:
                outputs[root] = {}

            current_output = outputs[root]

            if path.endswith(".png"):
                image = file_obj.read()

                current_output["img_bytes"] = image
            else:
                df = pd.read_csv(
                    BytesIO(file_obj.read()),
                    header=None,
                    names=["feature", "value", "num_categories"],
                )

                for index, row in df.iterrows():
                    current_output[row.feature] = row.value
                    current_output[f"{row.feature}_num_categories"] = row.num_categories

            if "img_bytes" in current_output and len(current_output) > 1:
                yield root, current_output
                del outputs[root]
                root = None

        if len(outputs) > 0:
            raise ValueError(
                f"Unable to extract the following samples: {list(outputs)}"
            )