Class DatasetFunc

monai/data/dataset.py:112–159

Execute function on the input dataset and leverage the output to act as a new Dataset. It can be used to load / fetch the basic dataset items, like the list of `image, label` paths. Or chain together to execute more complicated logic, like `partition_dataset`, `resample_datalist`, etc.

class DatasetFunc(Dataset):
    """
    Execute function on the input dataset and leverage the output to act as a new Dataset.
    It can be used to load / fetch the basic dataset items, like the list of `image, label` paths.
    Or chain together to execute more complicated logic, like `partition_dataset`, `resample_datalist`, etc.
    The `data` arg of `Dataset` will be applied to the first arg of callable `func`.
    Usage example::

        data_list = DatasetFunc(
            data="path to file",
            func=monai.data.load_decathlon_datalist,
            data_list_key="validation",
            base_dir="path to base dir",
        )
        # partition dataset for every rank
        data_partition = DatasetFunc(
            data=data_list,
            func=lambda **kwargs: monai.data.partition_dataset(**kwargs)[torch.distributed.get_rank()],
            num_partitions=torch.distributed.get_world_size(),
        )
        dataset = Dataset(data=data_partition, transform=transforms)

    Args:
        data: input data for the func to process, will apply to `func` as the first arg.
        func: callable function to generate dataset items.
        kwargs: other arguments for the `func` except for the first arg.

    """

    def __init__(self, data: Any, func: Callable, **kwargs) -> None:
        super().__init__(data=None, transform=None)  # type: ignore
        self.src = data
        self.func = func
        self.kwargs = kwargs
        self.reset()

    def reset(self, data: Any | None = None, func: Callable | None = None, **kwargs):
        """
        Reset the dataset items with specified `func`.

        Args:
            data: if not None, execute `func` on it, default to `self.src`.
            func: if not None, execute the `func` with specified `kwargs`, default to `self.func`.
            kwargs: other arguments for the `func` except for the first arg.

        """
        src = self.src if data is None else data
        self.data = self.func(src, **self.kwargs) if func is None else func(src, **kwargs)


class PersistentDataset(Dataset):