ornithoscope.py

import json
import os

import datasets
from datasets.tasks import ImageClassification


logger = datasets.logging.get_logger(__name__)


_CITATION = """"""

_DESCRIPTION = """\
Ornithoscope dataset is the dataset used to train the model for the Ornithoscope project.
"""

_HOMEPAGE = ""

_FOLDERS = [
    'iNatv1',
    'iNatv2',
    'PhotoFeederv1/task_05-01-2021',
]


class OrnithoscopeConfig(datasets.BuilderConfig):
    """BuilderConfig for Ornithoscope."""

    def __init__(
        self,
        classes: list[str],
        train_json: str,
        validation_json: str,
        test_json: str,
        **kwargs
    ):
        """BuilderConfig for Ornithoscope.

        Args:
            classes: list of classes.
            train_json: path to the json file containing the train annotations.
            validation_json: path to the json file containing the validation annotations.
            test_json: path to the json file containing the test annotations.
            **kwargs: keyword arguments forwarded to super.
        """
        super().__init__(version=datasets.Version("1.0.0"), **kwargs)
        self.classes = classes


class Ornithoscope(datasets.GeneratorBasedBuilder):

    VERSION = datasets.Version("1.0.0")
    BUILDER_CONFIGS = [
        OrnithoscopeConfig(
            name="DS3",
            description="The main dataset.",
            classes=[],  # TODO
            train_json="sets/DS3_train.json",
            validation_json="sets/DS3_val.json",
            test_json="sets/DS3_test.json",
        ),
    ]

    def _info(self) -> datasets.DatasetInfo:
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "image": datasets.Image(),
                    "label": datasets.ClassLabel(names=self.config.classes),
                }
            ),
            homepage=_HOMEPAGE,
            citation=_CITATION,
            task_templates=[ImageClassification(
                image_column="image",
                label_column="label",
            )],
        )

    def _split_generators(self, dl_manager: datasets.DownloadManager) -> list[datasets.SplitGenerator]:
        """Returns SplitGenerators."""
        archives = self._get_archives(dl_manager)

        # Get train paths.
        train_json = json.load(
            open(dl_manager.download_and_extract(self.config.train_json), 'r'))
        train_vals = []
        for id_path, value in train_json.items():
            root, file = os.path.split(id_path)
            path = os.path.join(archives[root], file)
            val = {
                "id_path": id_path,
                "path": path,
                "boxes": value['boxes'],
                "size": value['size'],
            }
            train_vals.append(val)

        # Get validation paths.
        validation_json = json.load(
            open(dl_manager.download_and_extract(self.config.validation_json), 'r'))
        validation_vals = []
        for id_path, value in validation_json.items():
            root, file = os.path.split(id_path)
            path = os.path.join(archives[root], file)
            val = {
                "id_path": id_path,
                "path": path,
                "boxes": value['boxes'],
                "size": value['size'],
            }
            validation_vals.append(val)

        # Get test paths.
        test_json = json.load(
            open(dl_manager.download_and_extract(self.config.test_json), 'r'))
        test_vals = []
        for id_path, value in test_json.item():
            root, file = os.path.split(id_path)
            path = os.path.join(archives[root], file)
            val = {
                "id_path": id_path,
                "path": path,
                "boxes": value['boxes'],
                "size": value['size'],
            }

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "values": train_vals,
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "values": validation_vals,
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "values": test_vals,
                },
            ),
        ]

    def _generate_examples(self, values: list) -> tuple:
        """Yields examples."""
        idx = 0
        for val in values:
            example = {
                "id_path": val["id_path"],
                "path": val["path"],
                "boxes": val["boxes"],
                "size": val["size"],
            }
            yield idx, example
            idx += 1

    def _get_archives(self, dl_manager: datasets.DownloadManager) -> dict:
        """Get the archives containing the images."""
        archives = {}
        for folder in _FOLDERS:
            i = 0
            while True:
                try:
                    archives[folder] = dl_manager.download_and_extract(
                        f'data/{folder}.tar'
                    )
                    i += 1
                except:
                    break
        return archives