dgp.utils.structures package¶
dgp.utils.structures.bounding_box_2d module¶
- class dgp.utils.structures.bounding_box_2d.BoundingBox2D(box, class_id=1, instance_id=None, color=(0, 0, 0), attributes=None, mode='ltwh')¶
Bases:
object2D bounding box object.
- box: np.ndarray[np.float32]
Array of 4 floats describing bounding box coordinates. Can be either ([l, t, r, b] or [l, t, w, h]).
- class_id: int, default: GENERIC_OBJECT_CLASS
Integer class ID (0 reserved for background).
- instance_id: int, default: None
Unique instance ID for bounding box. If None provided, the ID is a hash of the bounding box location and class.
- color: tuple, default: (0, 0, 0)
RGB tuple for bounding box color
- attributes: dict, default: None
Dictionary of attributes associated with bounding box. If None provided, defaults to empty dict.
- mode: str, default: ltwh
One of “ltwh” or “ltrb”. Corresponds to “[left, top, width, height]” representation or “[left, top, right, bottom]”
- Exception
Raised if the value of mode is unsupported.
- property area¶
- property attributes¶
- property class_id¶
- property color¶
- property hexdigest¶
- property instance_id¶
- intersection_over_union(other)¶
Compute intersection over union of this box against other(s).
- other: BoundingBox2D
A separate BoundingBox2D instance to compute IoU against.
- NotImplementedError
Unconditionally
- property ltrb¶
- property ltwh¶
- render(image)¶
Render bounding boxes on an image.
- image: PIL.Image or np.ndarray
Background image on which boxes are rendered
- image: PIL.Image or np.ndarray
Image with boxes rendered
- to_proto()¶
Serialize bounding box to proto object.
NOTE: Does not serialize class or instance information, just box geometry. To serialize a complete annotation, see dgp/annotations/bounding_box_2d_annotation.py
- BoundingBox2D.pb2
As defined in proto/annotations.proto
dgp.utils.structures.bounding_box_3d module¶
- class dgp.utils.structures.bounding_box_3d.BoundingBox3D(pose, sizes=array([0., 0., 0.], dtype=float32), class_id=1, instance_id=None, color=(0, 0, 0), attributes=None, num_points=0, occlusion=0, truncation=0.0, feature_ontology_type=None, sample_idx=None)¶
Bases:
object3D bounding box (cuboid) that is centered at pose with extent sizes.
- pose: dgp.utils.pose.Pose, (default: Pose())
Pose of the center of the 3D cuboid.
- sizes: np.float32, (default: np.float32([0,0,0]))
Extents of the cuboid (width, length, height).
- class_id: int, default: GENERIC_OBJECT_CLASS
Integer class ID (0 reserved for background).
- instance_id: int, default: None
Unique instance ID for bounding box. If None provided, the ID is a hash of the bounding box location and class.
- color: tuple, default: (0, 0, 0)
RGB tuple for bounding box color.
- attributes: dict, default: None
Dictionary of attributes associated with bounding box. If None provided, defaults to empty dict.
- num_points: int, default: 0
Number of LIDAR points associated with this bounding box.
- occlusion: int, default: 0
Occlusion state (KITTI3D style).
- truncation: float, default: 0
Fraction of truncation of object (KITTI3D style).
- feature_ontology_type: dgp.proto.features.FeatureType
Type of feature of attributions.
- sample_idx: int
index of sample in scene
- property attributes¶
- property class_id¶
- property corners¶
Get 8 corners of the 3D bounding box. Note: The pose is oriented such that x-forward, y-left, z-up. This corresponds to L (length) along x, W (width) along y, and H (height) along z.
- corners: np.ndarray (8 x 3)
Corners of the 3D bounding box.
- property edges¶
Get the 12 edge links of the 3D bounding box, indexed by the corners defined by self.corners`.
- edges: np.ndarray (12 x 2)
Edge links of the 3D bounding box.
- property feature_ontology_type¶
- property hexdigest¶
- property instance_id¶
- property num_points¶
- property occlusion¶
- property pose¶
- render(image, camera, line_thickness=2, class_name=None, font_scale=0.5)¶
Render the bounding box on the image.
- image: np.ndarray
Image (H, W, C) to render the bounding box onto. We assume the input image is in RGB format. The type must be uint8.
- camera: dgp.utils.camera.Camera
Camera used to render the bounding box.
- line_thickness: int, optional
Thickness of bounding box lines. Default: 2.
- class_name: str, optional
Class name of the bounding box. Default: None.
- font_scale: float, optional
Font scale used in text labels. Default: 0.5.
- image: np.uint8 array
Rendered image (H, W, 3).
- ValueError
Raised if image is not a 3-channel uint8 numpy array.
- TypeError
Raised if camera is not an instance of Camera.
- property sample_idx¶
- property sizes¶
- to_proto()¶
Serialize bounding box to proto object.
NOTE: Does not serialize class or instance information, just box properties. To serialize a complete annotation, see dgp/annotations/bounding_box_3d_annotation.py
- BoundingBox3D.pb2
As defined in proto/annotations.proto.
- property truncation¶
- property vectorize¶
Get a np.ndarray of with 10 dimensions representing the 3D bounding box
- box_3d: np.float32 array
Box with coordinates (pose.quat.qw, pose.quat.qx, pose.quat.qy, pose.quat.qz, pose.tvec.x, pose.tvec.y, pose.tvec.z, width, length, height).
dgp.utils.structures.instance_mask module¶
- class dgp.utils.structures.instance_mask.InstanceMask2D(mask, class_id=1, instance_id=None, color=(0, 0, 0), attributes=None)¶
Bases:
object2D instance mask object.
- mask: np.ndarray[bool, np.uint8, np.int64]
2D boolean array describiing instance mask.
- class_id: int, default: GENERIC_OBJECT_CLASS
Integer class ID (0 reserved for background).
- instance_id: int, default: None
Unique instance ID for instance mask. If None provided, the ID is a hash of the instance mask, location and class.
- attributes: dict, default: None
Dictionary of attributes associated with instance mask. If None provided, defaults to empty dict.
- property area¶
Compute intersection over union of this box against other(s).
- property attributes¶
- property bitmask¶
- property class_id¶
- property color¶
- property hexdigest¶
- property instance_id¶
- intersection_over_union(other)¶
Compute intersection over union of this box against other(s).
- other: InstanceMask2D
Another instance of InstanceMask2D to compute IoU against.
- NotImplementedError
Unconditionally.
- render(image)¶
Render instance masks on an image.
- image: PIL.Image or np.ndarray
Background image on which boxes are rendered
- image: PIL.Image or np.ndarray
Image with boxes rendered
- property rle¶
- class dgp.utils.structures.instance_mask.RLEMask(size, counts)¶
Bases:
objectContainer of RLE-encoded mask.
See https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocotools/mask.py for RLE format.
- size: list[int] or np.ndarray[int]
Height and width of mask.
- counts: list[int]
Count-encoding of RLE format.
- to_dict()¶
dgp.utils.structures.key_point_2d module¶
- class dgp.utils.structures.key_point_2d.KeyPoint2D(point, class_id=1, instance_id=None, color=(0, 0, 0), attributes=None)¶
Bases:
object2D keypoint object.
- point: np.ndarray[np.float32]
Array of 2 floats describing keypoint coordinates [x, y].
- class_id: int, default: GENERIC_OBJECT_CLASS
Integer class ID (0 reserved for background).
- instance_id: int, default: None
Unique instance ID for keypoint. If None provided, the ID is a hash of the keypoint location and class.
- color: tuple, default: (0, 0, 0)
RGB tuple for keypoint color
- attributes: dict, default: None
Dictionary of attributes associated with keypoint. If None provided, defaults to empty dict.
- property attributes¶
- property class_id¶
- property color¶
- property hexdigest¶
- property instance_id¶
- to_proto()¶
Serialize keypoint to proto object.
NOTE: Does not serialize class or instance information, just point geometry. To serialize a complete annotation, see dgp/annotations/key_point_2d_annotation.py
- KeyPoint2D.pb2
As defined in proto/annotations.proto
- property xy¶