Method track_transform_meta

monai/transforms/inverse.py:170–302

Update a stack of applied/pending transforms metadata of ``data``. Args: data: dictionary of data or `MetaTensor`. key: if data is a dictionary, data[key] will be modified. sp_size: the expected output spatial size when the transform is applied.

(
        cls,
        data,
        key: Hashable = None,
        sp_size=None,
        affine=None,
        extra_info: dict | None = None,
        orig_size: tuple | None = None,
        transform_info=None,
        lazy=False,
    )

    @classmethod
    def track_transform_meta(
        cls,
        data,
        key: Hashable = None,
        sp_size=None,
        affine=None,
        extra_info: dict | None = None,
        orig_size: tuple | None = None,
        transform_info=None,
        lazy=False,
    ):
        """
        Update a stack of applied/pending transforms metadata of ``data``.

        Args:
            data: dictionary of data or `MetaTensor`.
            key: if data is a dictionary, data[key] will be modified.
            sp_size: the expected output spatial size when the transform is applied.
                it can be tensor or numpy, but will be converted to a list of integers.
            affine: the affine representation of the (spatial) transform in the image space.
                When the transform is applied, meta_tensor.affine will be updated to ``meta_tensor.affine @ affine``.
            extra_info: if desired, any extra information pertaining to the applied
                transform can be stored in this dictionary. These are often needed for
                computing the inverse transformation.
            orig_size: sometimes during the inverse it is useful to know what the size
                of the original image was, in which case it can be supplied here.
            transform_info: info from self.get_transform_info().
            lazy: whether to push the transform to pending_operations or applied_operations.

        Returns:

            For backward compatibility, if ``data`` is a dictionary, it returns the dictionary with
            updated ``data[key]``. Otherwise, this function returns a MetaObj with updated transform metadata.
        """
        data_t = data[key] if key is not None else data  # compatible with the dict data representation
        out_obj = MetaObj()
        # after deprecating metadict, we should always convert data_t to metatensor here
        if isinstance(data_t, MetaTensor):
            out_obj.copy_meta_from(data_t, keys=out_obj.__dict__.keys())

        if lazy and (not get_track_meta()):
            warnings.warn("metadata is not tracked, please call 'set_track_meta(True)' if doing lazy evaluation.")

        if not lazy and affine is not None and isinstance(data_t, MetaTensor):
            # not lazy evaluation, directly update the metatensor affine (don't push to the stack)
            orig_affine = data_t.peek_pending_affine()
            orig_affine = convert_to_dst_type(orig_affine, affine, dtype=torch.float64)[0]
            try:
                affine = orig_affine @ to_affine_nd(len(orig_affine) - 1, affine, dtype=torch.float64)
            except RuntimeError as e:
                if orig_affine.ndim > 2:
                    if data_t.is_batch:
                        msg = "Transform applied to batched tensor, should be applied to instances only"
                    else:
                        msg = "Mismatch affine matrix, ensured that the batch dimension is not included in the calculation."
                    raise RuntimeError(msg) from e
                else:
                    raise