Corrects the calibration of KITTI's HDL-64 scan

Computes a matrix product for a batch of matrices Parameters ---------- a : torch.Tensor [..., N, M] b : torch.Tensor [..., M, K

A Decorator which defines the fields of a Dataclass from the attributes of a pyct_icp object Note: It should be added before the decorator @d

Aligns the target Point Cloud on the reference Point Cloud Args: ref_points (torch.Tensor): The reference points tensor

Aligns the target Point Cloud on the reference Point Cloud Args: ref_points (torch.Tensor): The reference points tensor

A Decorator for batched numpy or pytorch operator Which extends arrays in the first dimension to match the desired input shapes

Builds an elevation image

Builds a Spherical Image from a Point Cloud in place Parameters ---------- t_point_cloud: torch.Tensor [B, N, C >= 3] The fi

A Decorator for batched numpy unary operator Which checks the size of array against desired shapes

Overrides pytorch default collate function, to keep numpy arrays in dictionaries If `batch` is a dictionary, every key containing the key `n

Solve the Linear Least Square estimation A * x = b Args: x0 (torch.Tensor): The initial set of parameters `(B, D,)`

Computes the 3D Rigid transform associated to feature based 2D alignment

Computes the neighborhood of a given image tensor. More precisely, it extracts for each pixel location p, the K * K pixels in the nei

The Standard cost associated to the residuals

Huber cost computed from the residuals

Returns the weighted squared residuals

Returns the weighted squared residuals

Returns the weighted squared residuals

Returns the weighted squared residuals

Build the cross operator from a tensor of 3D vectors It is the 3x3 matrix A computed from a which verifies, for each vector b, A * b = a

Extracts the depth from a point cloud Parameters ---------- t_point_cloud : (torch.Tensor, np.ndarray) [B, N, 3] of [N, 3] A

Registers a new frame to the Map Note: `CT_ICP` requires frames with timestamps. If no timestamps are foun

Processes a new frame Estimates the motion for the new frame, and update the states of the different components (Local Map,

Registers the new frame

Applies all filters sequentially

Computes the exponential weighing of the losses in list_losses Parameters ---------- list_losses : list

Returns the computed relative poses along the trajectory

Returns: (train_dataset, eval_dataset, test_dataset) A tuple of `DatasetOfSequences` consisting of concatenated datasets

Returns the total elapsed time in calling process_next_frame

Returns the relative ground truth poses associated to a sequence of KITTI-360

Returns the ground truth poses associated to a sequence of KITTI's odometry benchmark

Returns the ground truth poses

No ground truth can be read from the ROSBAG

Returns the ground truth poses associated to a sequence of KITTI's odometry benchmark

Returns the last frame registered in the Local Map

Returns the last pointcloud registered to the local map

Returns the last pointcloud registered to the local map

Returns the estimated relative poses for the current sequence

Returns the Point-to-Plane residual closure The returned closure takes input a pose matrix or pose params tensor, And return

Returns the Point-to-Point residual jacobian closure The returned closure takes input a pose matrix or pose params tensor, A

Initializes the Algorithm ()

Sets the predicted motion as the identity systematically

Cleans and Initializes the Loop Closure Algorithm

Initialize/ReInitialize the state of the Algorithm and its components

Initialize/ReInitialize the state of the Algorithm and its components

Initializes the Odometry algorithm Clears the persisted relative poses, reset the _iter to 0 And loads the module parameters

Initialize the Local Map by releasing all persisted tensors

An initialization procedure called at the start of each sequence