spark_plug_anomaly.py

import os
from datasets import DatasetInfo, GeneratorBasedBuilder, SplitGenerator, Split, Features, ClassLabel, Image

class SparkPlugAnomaly(GeneratorBasedBuilder):
    def _info(self):
        return DatasetInfo(
            description="Spark Plug Visual Anomaly Detection Dataset for Edge Impulse FOMO-AD",
            features=Features({
                "image": Image(),
                "label": ClassLabel(names=["normal", "anomaly"])
            }),
            supervised_keys=("image", "label"),
        )

    def _split_generators(self, dl_manager):
        data_dir = os.path.join(dl_manager.manual_dir, "data")
        return [
            SplitGenerator(name=Split.TRAIN, gen_kwargs={"data_dir": data_dir}),
        ]

    def _generate_examples(self, data_dir):
        idx = 0
        for label in sorted(os.listdir(data_dir)):
            label_path = os.path.join(data_dir, label)
            if not os.path.isdir(label_path):
                continue
            for fname in os.listdir(label_path):
                if fname.lower().endswith((".jpg", ".jpeg", ".png")):
                    yield idx, {
                        "image": os.path.join(label_path, fname),
                        "label": label
                    }
                    idx += 1