cpas_method.py

cpas_toolbox.cpas_method

This module defines the interface for categorical pose and shape estimation methods.

This module defines two classes: PredictionDict and CPASMethod. PredictionDict defines the prediction produced by a CPASMethod. CPASMethod defines the interface used to evaluate categorical pose and shape estimation methods.

PredictionDict

Bases: TypedDict

Pose and shape prediction.

ATTRIBUTE	DESCRIPTION
position	Position of object center in camera frame. OpenCV convention. Shape (3,). TYPE: Tensor
orientation	Orientation of object in camera frame. OpenCV convention. Scalar-last quaternion, shape (4,). TYPE: Tensor
extents	Bounding box side lengths, shape (3,). TYPE: Tensor
reconstructed_pointcloud	Metrically-scaled reconstructed pointcloud in object frame. None if method does not perform reconstruction. TYPE: Optional[Tensor]
reconstructed_mesh	Metrically-scaled reconstructed mesh in object frame. None if method does not perform reconstruction. TYPE: Optional[TriangleMesh]

Source code in cpas_toolbox/cpas_method.py

class PredictionDict(TypedDict):
    """Pose and shape prediction.

    Attributes:
        position:
            Position of object center in camera frame. OpenCV convention. Shape (3,).
        orientation:
            Orientation of object in camera frame. OpenCV convention.
            Scalar-last quaternion, shape (4,).
        extents:
            Bounding box side lengths, shape (3,).
        reconstructed_pointcloud:
            Metrically-scaled reconstructed pointcloud in object frame.
            None if method does not perform reconstruction.
        reconstructed_mesh:
            Metrically-scaled reconstructed mesh in object frame.
            None if method does not perform reconstruction.
    """

    position: torch.Tensor
    orientation: torch.Tensor
    extents: torch.Tensor
    reconstructed_pointcloud: Optional[torch.Tensor]
    reconstructed_mesh: Optional[o3d.geometry.TriangleMesh]

CPASMethod

Bases: ABC

Interface class for categorical pose and shape estimation methods.

Source code in cpas_toolbox/cpas_method.py

class CPASMethod(ABC):
    """Interface class for categorical pose and shape estimation methods."""

    def __init__(self, config: dict, camera: camera_utils.Camera) -> None:
        """Initialize categorical pose and shape estimation method.

        Args:
            config: Method configuration dictionary.
            camera: Camera used for the input image.
        """
        pass

    def inference(
        self,
        color_image: torch.Tensor,
        depth_image: torch.Tensor,
        instance_mask: torch.Tensor,
        category_str: str,
    ) -> PredictionDict:
        """Run a method to predict pose and shape of an object.

        Args:
            color_image: The color image, shape (H, W, 3), RGB, 0-1, float.
            depth_image: The depth image, shape (H, W), meters, float.
            instance_mask: Mask of object of interest. (H, W), bool.
            category_str: The category of the object.
        """
        pass

__init__

__init__(config: dict, camera: camera_utils.Camera) -> None

Initialize categorical pose and shape estimation method.

PARAMETER	DESCRIPTION
config	Method configuration dictionary. TYPE: dict
camera	Camera used for the input image. TYPE: Camera

Source code in cpas_toolbox/cpas_method.py

def __init__(self, config: dict, camera: camera_utils.Camera) -> None:
    """Initialize categorical pose and shape estimation method.

    Args:
        config: Method configuration dictionary.
        camera: Camera used for the input image.
    """
    pass

inference

inference(
    color_image: torch.Tensor,
    depth_image: torch.Tensor,
    instance_mask: torch.Tensor,
    category_str: str,
) -> PredictionDict

Run a method to predict pose and shape of an object.

PARAMETER	DESCRIPTION
color_image	The color image, shape (H, W, 3), RGB, 0-1, float. TYPE: Tensor
depth_image	The depth image, shape (H, W), meters, float. TYPE: Tensor
instance_mask	Mask of object of interest. (H, W), bool. TYPE: Tensor
category_str	The category of the object. TYPE: str

Source code in cpas_toolbox/cpas_method.py

def inference(
    self,
    color_image: torch.Tensor,
    depth_image: torch.Tensor,
    instance_mask: torch.Tensor,
    category_str: str,
) -> PredictionDict:
    """Run a method to predict pose and shape of an object.

    Args:
        color_image: The color image, shape (H, W, 3), RGB, 0-1, float.
        depth_image: The depth image, shape (H, W), meters, float.
        instance_mask: Mask of object of interest. (H, W), bool.
        category_str: The category of the object.
    """
    pass