Function get_extreme_points

monai/transforms/utils.py:1580–1635

Generate extreme points from an image. These are used to generate initial segmentation for annotation models. An optional perturbation can be passed to simulate user clicks. Args: img: Image to generate extreme points from. Expected Shape is ``(spatial_dim1, [, spat

(
    img: NdarrayOrTensor, rand_state: np.random.RandomState | None = None, background: int = 0, pert: float = 0.0
)

def get_extreme_points(
    img: NdarrayOrTensor, rand_state: np.random.RandomState | None = None, background: int = 0, pert: float = 0.0
) -> list[tuple[int, ...]]:
    """
    Generate extreme points from an image. These are used to generate initial segmentation
    for annotation models. An optional perturbation can be passed to simulate user clicks.

    Args:
        img:
            Image to generate extreme points from. Expected Shape is ``(spatial_dim1, [, spatial_dim2, ...])``.
        rand_state: `np.random.RandomState` object used to select random indices.
        background: Value to be consider as background, defaults to 0.
        pert: Random perturbation amount to add to the points, defaults to 0.0.

    Returns:
        A list of extreme points, its length is equal to 2 * spatial dimension of input image.
        The output format of the coordinates is:

        [1st_spatial_dim_min, 1st_spatial_dim_max, 2nd_spatial_dim_min, ..., Nth_spatial_dim_max]

    Raises:
        ValueError: When the input image does not have any foreground pixel.
    """
    check_non_lazy_pending_ops(img, name="get_extreme_points")
    if rand_state is None:
        rand_state = np.random.random.__self__  # type: ignore
    indices = where(img != background)
    if np.size(indices[0]) == 0:
        raise ValueError("get_extreme_points: no foreground object in mask!")

    def _get_point(val, dim):
        """
        Select one of the indices within slice containing val.

        Args:
            val : value for comparison
            dim : dimension in which to look for value
        """
        idx = where(indices[dim] == val)[0]
        idx = idx.cpu() if isinstance(idx, torch.Tensor) else idx
        idx = rand_state.choice(idx) if rand_state is not None else idx
        pt = []
        for j in range(img.ndim):
            # add +- pert to each dimension
            val = int(indices[j][idx] + 2.0 * pert * (rand_state.rand() if rand_state is not None else 0.5 - 0.5))
            val = max(val, 0)
            val = min(val, img.shape[j] - 1)
            pt.append(val)
        return pt

    points = []
    for i in range(img.ndim):
        points.append(tuple(_get_point(indices[i].min(), i)))
        points.append(tuple(_get_point(indices[i].max(), i)))

    return points