TeTIm-Eval.py

import os
import json
import datasets

logger = datasets.logging.get_logger(__name__)

_CITATION = """\
TODO
"""

_HOMEPAGE = ""

_DESCRIPTION = """\
Text To Image Evaluation (TeTIm-Eval)
"""

_URLS = {
    "captioned": "https://huggingface.co/datasets/galatolo/TeTIm-Eval/resolve/main/data/TeTIm-Eval-Mini.zip",
    "uncaptioned": "https://huggingface.co/datasets/galatolo/TeTIm-Eval/resolve/main/data/TeTIm-Eval.zip"
}

_CLASSES = [
    "digital_art",
    "sketch_art",
    "traditional_art",
    "baroque_painting",
    "high_renaissance_painting",
    "neoclassical_painting",
    "animal_photo",
    "food_photo",
    "landscape_photo",
    "person_photo"
]

_CATEGORIES = [
    "art",
    "painting",
    "photo"
]

_MAP_CATEGORY = {
    _CLASSES[0]: _CATEGORIES[0],
    _CLASSES[1]: _CATEGORIES[0],
    _CLASSES[2]: _CATEGORIES[0],

    _CLASSES[3]: _CATEGORIES[1],
    _CLASSES[4]: _CATEGORIES[1],
    _CLASSES[5]: _CATEGORIES[1],

    _CLASSES[6]: _CATEGORIES[2],
    _CLASSES[7]: _CATEGORIES[2],
    _CLASSES[8]: _CATEGORIES[2],
    _CLASSES[9]: _CATEGORIES[2],
}

_FOLDERS = {
    "captioned": {
        _CLASSES[0]: "TeTIm-Eval-Mini/sampled_art_digital",
        _CLASSES[1]: "TeTIm-Eval-Mini/sampled_art_sketch",
        _CLASSES[2]: "TeTIm-Eval-Mini/sampled_art_traditional",
        _CLASSES[3]: "TeTIm-Eval-Mini/sampled_painting_baroque",
        _CLASSES[4]: "TeTIm-Eval-Mini/sampled_painting_high-renaissance",
        _CLASSES[5]: "TeTIm-Eval-Mini/sampled_painting_neoclassicism",
        _CLASSES[6]: "TeTIm-Eval-Mini/sampled_photo_animal",
        _CLASSES[7]: "TeTIm-Eval-Mini/sampled_photo_food",
        _CLASSES[8]: "TeTIm-Eval-Mini/sampled_photo_landscape",
        _CLASSES[9]: "TeTIm-Eval-Mini/sampled_photo_person",
    },
    "uncaptioned": {
        _CLASSES[0]: "TeTIm-Eval/sampled_art_digital",
        _CLASSES[1]: "TeTIm-Eval/sampled_art_sketch",
        _CLASSES[2]: "TeTIm-Eval/sampled_art_traditional",
        _CLASSES[3]: "TeTIm-Eval/sampled_painting_baroque",
        _CLASSES[4]: "TeTIm-Eval/sampled_painting_high-renaissance",
        _CLASSES[5]: "TeTIm-Eval/sampled_painting_neoclassicism",
        _CLASSES[6]: "TeTIm-Eval/sampled_photo_animal",
        _CLASSES[7]: "TeTIm-Eval/sampled_photo_food",
        _CLASSES[8]: "TeTIm-Eval/sampled_photo_landscape",
        _CLASSES[9]: "TeTIm-Eval/sampled_photo_person",
    }
}

class TeTImConfig(datasets.BuilderConfig):
    def __init__(self, **kwargs):
        super(TeTImConfig, self).__init__(**kwargs)


class TeTIm(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        TeTImConfig(
            name="captioned",
            version=datasets.Version("1.0.0", ""),
            description="A random sampling of 300 text-images pairs (30 per category) from the TeTIm dataset, manually annotated by the same person",
        ),
        TeTImConfig(
            name="uncaptioned",
            version=datasets.Version("1.0.0", ""),
            description="2500 labelled images (250 per category) from the TeTIm dataset",
        ),
    ]

    DEFAULT_CONFIG_NAME="captioned"

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "id": datasets.Value("int32"),
                    "image": datasets.Image(),
                    "caption": datasets.Value("string"),
                    "category": datasets.ClassLabel(num_classes=len(_CATEGORIES), names=_CATEGORIES),
                    "class": datasets.ClassLabel(num_classes=len(_CLASSES), names=_CLASSES)
                }
            ),
            supervised_keys=None,
            homepage=_HOMEPAGE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        target = os.environ.get(f"TETIMEVAL_{self.config.name}", _URLS[self.config.name])
        downloaded_files = dl_manager.download_and_extract(target)

        return [
            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"path": downloaded_files}),
        ]

    
    def _generate_examples(self, path):
        id = 0
        for _class, folder in _FOLDERS[self.config.name].items():
            images_folder = os.path.join(path, folder, "images")
            annotations_folder = os.path.join(path, folder, "annotations")

            for image in os.listdir(images_folder):
                image_id = int(image.split(".")[0])
                annotation_file = os.path.join(annotations_folder, f"{image_id}.json")
                with open(annotation_file) as f:
                    annotation = json.load(f)
                
                yield id, {
                    "id": id,
                    "image": os.path.join(images_folder, image),
                    "caption": annotation["caption"],
                    "category": _CATEGORIES.index(_MAP_CATEGORY[_class]),
                    "class": _CLASSES.index(_class)
                }
                id += 1


if __name__ == "__main__":
    from datasets import load_dataset
    dataset_config = {
        "LOADING_SCRIPT_FILES": os.path.join(os.getcwd(), "TeTIm-Eval.py"),
        "CONFIG_NAME": "uncaptioned",
    }
    ds = load_dataset(
        dataset_config["LOADING_SCRIPT_FILES"],
        dataset_config["CONFIG_NAME"],
    )
    print(ds)

    for i, e in zip(range(0, 10), ds["test"]):
        print(i, e)