bird-species-dataset.py

import os
import datasets
from datasets import SplitGenerator, Split, ImageClassification
from kaggle.api.kaggle_api_extended import KaggleApi

_CITATION = """\
@TECHREPORT{gpiosenka/100-bird-species,
    author = {gpiosenka},
    title = {BIRDS 525 SPECIES- IMAGE CLASSIFICATION},
    institution = {},
    year = {2023}
}
"""

_DESCRIPTION = """\
A dataset of bird species downloaded from kaggle. """

_HOMEPAGE = "https://www.kaggle.com/datasets/gpiosenka/100-bird-species/"

_DATA_DIR = 'data/'

def _CLASSES() -> list[str]:
    # reads from bird_labels.txt, line by line
    with open("birds_labels.txt") as f:
        return f.read().splitlines()

class BirdSpeciesDatasetDataset(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("0.1.0")

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="plain_text",
            version=datasets.Version("0.1.0", ""),
            description="Import of BIRDS 525 SPECIES Data Set",
        )
    ]

    def _info(self):
        _NAMES = _CLASSES()

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "image": datasets.Image(),
                    "label": datasets.features.ClassLabel(names=_NAMES),
                }
            ),
            supervised_keys=("image", "label"),
            homepage=_HOMEPAGE,
            citation=_CITATION,
            task_templates=ImageClassification(image_column="image", label_column="label"),
        )
    
    def _split_generators(self, dl_manager):
        data_dir = _DATA_DIR
        # Downloading the dataset
        if not os.path.exists(data_dir):
            kaggle_api = KaggleApi()
            kaggle_api.authenticate()
            kaggle_api.dataset_download_files('gpiosenka/100-bird-species', path=data_dir, unzip=True)

            # There is a bug in the dataset, where one of the folders is named "PARAKETT  AUKLET" instead of "PARAKETT AUKLET"
            # We fix it by adding a space to the valid folder, because train and test are wrong and also the names in the labels file
            # Fixing the path
            fault_path = os.path.join(data_dir, 'valid', 'PARAKETT AUKLET')
            correct_path = os.path.join(data_dir, 'valid', 'PARAKETT  AUKLET')
            os.rename(fault_path, correct_path)

        return [
            SplitGenerator(name=Split.TRAIN, gen_kwargs={"filepath": os.path.join(data_dir, "train")}),
            SplitGenerator(name=Split.VALIDATION, gen_kwargs={"filepath": os.path.join(data_dir, "valid")}),
            SplitGenerator(name=Split.TEST, gen_kwargs={"filepath": os.path.join(data_dir, "test")}),
        ]

    def _generate_examples(self, filepath):
        """Yields examples."""
        idx = 0
        for label in os.listdir(filepath):
            for f in os.listdir(os.path.join(filepath, label)):
                record = {
                    "image": os.path.join(filepath, label, f),
                    "label": label,
                }
                yield idx, record
                idx += 1

    


# To use the dataset:
# from datasets import load_dataset
# dataset = load_dataset('path/to/birds.py')