Initializes the ATrainer It will in order : - load the training, validation and test dataset_config - load t

Interpolates the poses from the ground truth using the provided timestamps Parameters: ground_truth (np.ndarray): The ground trut

Returns the inverse of the pose matrix Parameters ---------- params_tensor : [B, 6/7] or [B, 4, 4]

Returns whether the object contains any missing fields

Loads trajectory constraints from disk

Loads the Datasets

Returns the Point-to-Plane Loss

Returns the Point-to-Point Loss

Return the loss module used to train the model

Returns the nearest neighbors by projective data association Projects the points in the image plane

Returns the nearest neighbors, and associated normals computed with a KdTree

Returns the identity

The key (in data_dict) for xyz pointcloud

Generates a pixel grid of size [B, H, W, 2]

Computes a pillar hash for each voxel in an array of voxel indices A pillar hash is a hash for the first two coordinate of a voxel (x, y)

Reshapes a point cloud into a projection map Parameters ---------- points : torch.Tensor [K * h * w, num_channels] h : the heigh

Returns the PoseNet Prediction Module

Parameters ---------- pointcloud : torch.Tensor The point cloud tensor [B, N, 3] to project in the image plane

Project the pointcloud in the Spherical image Parameters ---------- pointcloud : torch.Tensor [B, N, K>=3] h

Default SphericalProjetor for UrbanLoco dataset (projection of a pointcloud into a Vertex Map)

Default SphericalProjetor for KITTI (projection of a pointcloud into a Vertex Map)

Default SphericalProjetor for KITTI (projection of a pointcloud into a Vertex Map)

Rescales the intrinsics parameters of the projection from the arguments Parameters ---------- new_height : int

The Spherical projection does not need to rescale its intrinsics parameters

Parameters ---------- new_height : int The new height of the projector new_width The new widt

Returns a rescaled Spherical projector

A Robust Kernel for Least-Square minimization

A Decorator which adds a set of defaults a dataclass

No actions required

Returns a key which defines an absolute constraint between poses

Returns the tuples (dataset_config, sequence_name) for train, eval and test split on FordCampus

Returns ------- (train_dataset, eval_dataset, test_dataset, transform) : tuple train_dataset : (list, list)

Returns ------- (train_dataset, eval_dataset, test_dataset, transform) : tuple train_dataset : (list, list)

Returns ------- (train_dataset, eval_dataset, test_dataset, transform) : tuple train_dataset : (list, list)

Returns ------- (train_dataset, eval_dataset, test_dataset, transform) : tuple train_dataset : (list,

Builds a new map from the aggregated pointcloud `pointcloud`

Updates the local map and (registers a new frame into it)

Updates the local map and (registers a new frame into it)

Returns a scalar weight of 1 for standard least square (to avoid) unnecessary computations

Returns the poses expressed in the world frame

Extracts the xyz fields of a point cloud Parameters ---------- t_point_cloud : A [B, N, C >= 3] or a [N, C >= 3] array Extra