Method __call__

monai/transforms/spatial/array.py:3202–3238

(
        self, image: NdarrayOrTensor, size: int | tuple[int, int] | np.ndarray | None = None
    )

        self.size = None if size is None else ensure_tuple_rep(size, len(self.grid))

    def __call__(
        self, image: NdarrayOrTensor, size: int | tuple[int, int] | np.ndarray | None = None
    ) -> list[NdarrayOrTensor]:
        input_size = self.size if size is None else ensure_tuple_rep(size, len(self.grid))

        if self.grid == (1, 1) and input_size is None:
            return [image]
        if isinstance(image, MetaTensor) and image.pending_operations:
            warnings.warn("MetaTensor img has pending operations, transform may return incorrect results.")
        split_size, steps = self._get_params(image.shape[1:], input_size)
        patches: list[NdarrayOrTensor]
        as_strided_func: Callable
        if isinstance(image, torch.Tensor):
            as_strided_func = torch.as_strided
            c_stride, x_stride, y_stride = image.stride()
        elif isinstance(image, np.ndarray):
            as_strided_func = np.lib.stride_tricks.as_strided
            c_stride, x_stride, y_stride = image.strides
        else:
            raise ValueError(f"Input type [{type(image)}] is not supported.")

        x_step, y_step = steps
        n_channels = image.shape[0]
        strided_image = as_strided_func(
            image,
            (*self.grid, n_channels, split_size[0], split_size[1]),
            (x_stride * x_step, y_stride * y_step, c_stride, x_stride, y_stride),
        )
        # Flatten the first two dimensions
        strided_image = strided_image.reshape(-1, *strided_image.shape[2:])
        # Make a list of contiguous patches
        if isinstance(image, torch.Tensor):
            patches = [p.contiguous() for p in strided_image]
        elif isinstance(image, np.ndarray):
            patches = [np.ascontiguousarray(p) for p in strided_image]

        return patches

    def _get_params(self, image_size: Sequence[int] | np.ndarray, size: Sequence[int] | np.ndarray | None = None):
        """