Associate to a pointcloud (given by a filename) the closest pose (in terms of timestamps) to the ground truth

Projects the points of a PointCloud tensor in the image plane Parameters ---------- pointcloud : torch.Tensor [B, N,

Returns the Default Spherical Image projector associated to the dataset_config

Reads a calibration file from KITTI KITTI's calibration files map keys to transform / projection / calibration matrices, Args:

Reads a ground truth file from KITTI odometry's benchmark Returns poses in the coordinate system of the left camera With the z axis faci

The key of the expected relative pose filled by the odometry at each frame The SLAM algorithm expects a [4, 4] numpy pose transform array und

Returns the Rosbag config for the given sequence

Saves metrics and trajectory in a folder on disk

Returns an array of colors from an array of values Parameters: values (np.ndarray): an array of scalars `(N,)` cmap (str): The na

Default method to send a dictionary to a device By default, only tensors are sent to the GPU Parameters ----------

Converts an image tensor to a np.array The image extracted from the tensor will obey the following rules, depending on the size of the tensor

Computes a spherical projection of the points of a point cloud It will compute the pixel values of the points in t_pointcloud Parameter

>>> check_tensor(torch_rx(torch.randn(6), torch.randn(6)), [6, 3, 3])

>>> check_tensor(torch_ry(torch.randn(6), torch.randn(6)), [6, 3, 3])

Adds a new item to the meter

Computes the normal distribution of points in each voxel Args: pointcloud (np.ndarray): The input pointcloud `(n, 3)` [np.float32]

Writes an array of poses to disk Parameters ---------- file_path : str poses : np.ndarray [N, 4, 4]

Returns an Image Visualizer based on OpenCV if the package cv2 was found

Agglomerate window data for a sequence Args: batch (List): A list of elements which are to be aggregated into a batch of

Computes the normal distribution of points in each voxel Args: pointcloud (np.ndarray): The input pointcloud `(n, 3)` [np.float32]

Compute inverses without division by det; ...xv3xc3 input, or array of matrices assumed Parameters ---------- tensor : torch.Tensor

Dot arrays of vecs; contract over last indices

Returns an initial pose from the dict

Inverse of a stack of 3x3 matrices Parameters ---------- tensor : torch.Tensor

Efficiently computes the inverse-transpose for stack of 3x3 matrices Parameters ---------- tensor : torch.Tensor or np.ndarray

Computes matrix to vector product for a batch of tensors Parameters ---------- a : torch.Tensor b : torch.Tensor

Reads and interprets the input from the data_dict

Regex for the search of relative constraints in the new frame dict

Computes the odometry metrics ATE, ARE, tr_err, rot_err for the sequence sequence_id, Saves the result in the log_dir, the trajectori

Aggregates the point clouds of the main topic

Aligns corresponding pair of 3D points Computes the optimal rigid transform between reference points (`ref_points`) and targ

Applies the rotation part of the pose on the point cloud or normal cloud Parameters ---------- tensor : [B, N, 3]

Builds the benchmark

Builds the image from the pointcloud (which will be matched by 2D feature based alignment)

Builds the model which allows to construct the nearest neighbor computation

Delete all previous data of the LoopClosure

Estimates via Gauss Newton the non linear least-square objective Args: x0 (torch.Tensor): The initial set of parameters

Aligns the images using corresponding pair of points (with potentially many outliers) By default, the best homography is fou

Compute ring ids by grouping points by polar angle bins (in spherical projection)

Computes the 3D transform from the aligned points

1x1 convolution with padding

Converts a [4, 4] pose tensor to the desired type Returns a tensor (either a numpy.ndarray or torch.Tensor depending on dest type) >>> check_

The Cost function associated to the Least Square Scheme

Returns the weighted squared residuals

Converts data to a Tensor for compatible data types Parameters ---------- data : An data to convert to torch The data can b