Class MetaObj

monai/data/meta_obj.py:63–244

Abstract base class that stores data as well as any extra metadata. This allows for subclassing `torch.Tensor` and `np.ndarray` through multiple inheritance. Metadata is stored in the form of a dictionary. Behavior should be the same as extended class (e.g., `torch.Tensor` or `np

class MetaObj:
    """
    Abstract base class that stores data as well as any extra metadata.

    This allows for subclassing `torch.Tensor` and `np.ndarray` through multiple inheritance.

    Metadata is stored in the form of a dictionary.

    Behavior should be the same as extended class (e.g., `torch.Tensor` or `np.ndarray`)
    aside from the extended meta functionality.

    Copying of information:

        * For `c = a + b`, then auxiliary data (e.g., metadata) will be copied from the
          first instance of `MetaObj` if `a.is_batch` is False
          (For batched data, the metadata will be shallow copied for efficiency purposes).

    """

    def __init__(self) -> None:
        self._meta: dict = MetaObj.get_default_meta()
        self._applied_operations: list = MetaObj.get_default_applied_operations()
        self._pending_operations: list = MetaObj.get_default_applied_operations()  # the same default as applied_ops
        self._is_batch: bool = False

    @staticmethod
    def flatten_meta_objs(*args: Iterable):
        """
        Recursively flatten input and yield all instances of `MetaObj`.
        This means that for both `torch.add(a, b)`, `torch.stack([a, b])` (and
        their numpy equivalents), we return `[a, b]` if both `a` and `b` are of type
        `MetaObj`.

        Args:
            args: Iterables of inputs to be flattened.
        Returns:
            list of nested `MetaObj` from input.
        """
        for a in itertools.chain(*args):
            if isinstance(a, (list, tuple)):
                yield from MetaObj.flatten_meta_objs(a)
            elif isinstance(a, MetaObj):
                yield a

    @staticmethod
    def copy_items(data):
        """returns a copy of the data. list and dict are shallow copied for efficiency purposes."""
        if is_immutable(data):
            return data
        if isinstance(data, (list, dict, np.ndarray)):
            return data.copy()
        if isinstance(data, torch.Tensor):
            return data.detach().clone()
        return deepcopy(data)

    def copy_meta_from(self, input_objs, copy_attr=True, keys=None):
        """
        Copy metadata from a `MetaObj` or an iterable of `MetaObj` instances.