l5kit.geometry package

l5kit.geometry.angle_between_vectors(v1: numpy.ndarray, v2: numpy.ndarray) → float

angle_between_vectors returns the angle in radians between two vectors.

Parameters

  • v1 (np.ndarray) – Vector 1 of shape (N)

  • v2 (np.ndarray) – Vector 2 of same shape as v1

Returns

angle in radians

Return type

float

l5kit.geometry.angular_distance(angle_a: Union[float, numpy.ndarray], angle_b: Union[float, numpy.ndarray]) → Union[float, numpy.ndarray]

A function that takes two arrays of angles in radian and compute the angular distance, wrap the angular distance such that they are always in the [-pi, pi) range.

Parameters

  • angle_a (np.ndarray, float) – first array of angles in radians

  • angle_b (np.ndarray, float) – second array of angles in radians

Returns

angular distance in radians between two arrays of angles

l5kit.geometry.compute_agent_pose(agent_centroid_m: numpy.ndarray, agent_yaw_rad: float) → numpy.ndarray

Return the agent pose as a 3x3 matrix. This corresponds to world_from_agent matrix.

Parameters

  • agent_centroid_m (np.ndarry) – 2D coordinates of the agent

  • agent_yaw_rad (float) – yaw of the agent

Returns

3x3 world_from_agent matrix

Return type

(np.ndarray)

l5kit.geometry.compute_yaw_around_north_from_direction(direction_vector: numpy.ndarray) → float

compute_yaw_from_direction computes the yaw as angle between a 2D input direction vector and the y-axis direction vector (0, 1).

Args:

direction_vector (np.ndarray): Vector of shape (2,)

Returns:

float: angle to (0,1) vector in radians

l5kit.geometry.crop_rectangle_from_image(image: numpy.ndarray, corners: numpy.ndarray) → numpy.ndarray

crop_rectangle_from_image takes an image and 4 corners in pixel coordinates, it returns the sub-image inside that cropped area, rotated upright.

Parameters

  • image (np.ndarray) – image to crop from

  • corners (np.ndarray) – corners, array of shape (4,2)

Returns

crop from input containing the corners

Return type

np.ndarray

l5kit.geometry.ecef_to_geodetic(point: Union[numpy.ndarray, Sequence[float]]) → numpy.ndarray

Convert given ECEF coordinate into latitude, longitude, altitude.

Parameters

point (Union[np.ndarray, Sequence[float]]) – ECEF coordinate vector

Returns

latitude, altitude, longitude

Return type

np.ndarray

l5kit.geometry.geodetic_to_ecef(lla_point: Union[numpy.ndarray, Sequence[float]]) → numpy.ndarray

Convert given latitude, longitude, and optionally altitude into ECEF coordinates. If no altitude is given, altitude 0 is assumed.

Parameters

lla_point (Union[np.ndarray, Sequence[float]]) – Latitude, Longitude and optionally Altitude

Returns

3D ECEF coordinate

Return type

np.ndarray

l5kit.geometry.normalize_intensity(x: numpy.ndarray, max_intensity: float) → numpy.ndarray

Normalize (divide by max) and clip intensity values to fall between 0 and 1.

Parameters

  • x (np.npdarray) – numpy array of any shape

  • max_intensity (float) – Maximum intensity value (anything above this will become 1)

Returns

np.ndarray – array of same type and shape as x with values between 0 and 1 only

l5kit.geometry.points_within_bounds(coords: numpy.ndarray, shape: Union[Collection[int], numpy.ndarray]) → numpy.ndarray
Parameters

  • coords (np.ndarray) – (N,3)-shaped array containing points.

  • shape (tuple of ints or np.ndarray) – shape to use as bounds, should be length 3

Returns

Binary mask for given coords array which is True for points that fall within the bounds of shape.

l5kit.geometry.rotation33_as_yaw(rotation: numpy.ndarray) → float

Compute the yaw component of given 3x3 rotation matrix.

Parameters

rotation (np.ndarray) – 3x3 rotation matrix (np.float64 dtype recommended)

Returns

yaw rotation in radians

Return type

float

l5kit.geometry.transform_point(point: numpy.ndarray, transf_matrix: numpy.ndarray) → numpy.ndarray
Transform a single vector using transformation matrix.

This function call transform_points internally

Parameters

  • point (np.ndarray) – vector of shape (N)

  • transf_matrix (np.ndarray) – transformation matrix of shape (N+1, N+1)

Returns

vector of same shape as input point

Return type

np.ndarray

l5kit.geometry.transform_points(points: numpy.ndarray, transf_matrix: numpy.ndarray) → numpy.ndarray

Transform a set of 2D/3D points using the given transformation matrix. Assumes row major ordering of the input points. The transform function has 3 modes: - points (N, F), transf_matrix (F+1, F+1)

all points are transformed using the matrix and the output points have shape (N, F).

  • points (B, N, F), transf_matrix (F+1, F+1)

    all sequences of points are transformed using the same matrix and the output points have shape (B, N, F). transf_matrix is broadcasted.

  • points (B, N, F), transf_matrix (B, F+1, F+1)

    each sequence of points is transformed using its own matrix and the output points have shape (B, N, F).

Note this function assumes points.shape[-1] == matrix.shape[-1] - 1, which means that last rows in the matrices do not influence the final results. For 2D points only the first 2x3 parts of the matrices will be used.

Parameters

  • points (np.ndarray) – Input points of shape (N, F) or (B, N, F)

  • F = 2 or 3 depending on input points are 2D or 3D points. (with) –

  • transf_matrix (np.ndarray) – Transformation matrix of shape (F+1, F+1) or (B, F+1, F+1) with F = 2 or 3.

Returns

Transformed points of shape (N, F) or (B, N, F) depending on the dimensions of the input points.

Return type

np.ndarray

l5kit.geometry.vertical_flip(tm: numpy.ndarray, y_dim_size: int) → numpy.ndarray

Return a new matrix that also performs a flip on the y axis.

Parameters

  • tm – the original 3x3 matrix

  • y_dim_size – this should match the resolution on y. It makes all coordinates positive

Returns: a new 3x3 matrix.

l5kit.geometry.voxel_coords_to_intensity_grid(voxel_coords: numpy.ndarray, shape: tuple, dtype: numpy.dtype = <class 'numpy.float32'>, drop_out_of_bounds: bool = True) → numpy.ndarray

Puts coords into a grid: for each grid cell the number of points is written there.

Parameters

  • voxel_coords (np.ndarray) – input array with coords (N,3) in intensity grid

  • shape (tuple of ints) – intensity grid shape

Keyword Arguments

  • dtype (data-type) – data type for the intensity grid (default: {np.float32})

  • drop_out_of_bounds (bool) – [description] (default: {True})

Returns

np.ndarray – Array with given shape, the value of each cell is the amount of coords for that point.

l5kit.geometry.yaw_as_rotation33(yaw: float) → numpy.ndarray

Create a 3x3 rotation matrix from given yaw. The rotation is counter-clockwise and it is equivalent to: [cos(yaw), -sin(yaw), 0.0], [sin(yaw), cos(yaw), 0.0], [0.0, 0.0, 1.0],

Parameters

yaw (float) – yaw rotation in radians

Returns

3x3 rotation matrix

Return type

np.ndarray