Advanced string replacement. Given a text: - replace all elements in src by the corresponding element in tgt - remove all elements in rm

Scales each track (N) independently by median norm ratio.

Associate two dictionaries of (stamp, data). As the time stamps never match exactly, we aim to find the closest match for every input tuple w

Build K-NN adjacency matrix with batch support. Args: points (torch.Tensor): (B, N, 3) k (int): Number of neighbors per point (e

Read camera data, return (M,N) tuple. M is the intrinsic matrix, N is the extrinsic matrix, so that x = M*N*X, with x being a point

Clean HuggingFace cache to force fresh download

Clean HuggingFace cache to force fresh download

Method: 1) express all 3d points in each camera coordinate frame 2) if they're in front of a depthmap --> then lower their confidence

Modify camera intrinsics to follow a different convention. Coordinates of the center of the top-left pixels are by default: - (0.5, 0.5)

Photometric augmentation

Args: x2d_proj: Shape (*, n, 2) x2d: Shape (*, n, 2) w2d: Shape (*, n, 2) jac_cam: Shape (*,

compute optical flow color map :param u: optical flow horizontal map :param v: optical flow vertical map :return: optical flow in col

Compute a HxWx3 pointmap in camera frame from a HxW depth map.

Compute a scale-invariant depth loss without taking logs. We: 1) Solve for alpha to best align pred_depth to gt_depth in an MSE sense.

Generate a normalized stride weight list representing the sampling probability of each stride in the current epoch. As the

Convert TUM RGB-D dataset to TapVid3D format.

Create a commandline string to generate a scene.

return PixelwiseTaskWithDPT for given net params

Read depth data from file, return as numpy array.

Args: - depthmap (HxW array): - camera_intrinsics: a 3x3 matrix Returns: pointmap of absolute coordinates (HxWx3 arra

dnames: list of datasets with image pairs, separated by +

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

run encoder for a pair of images it is actually ~5% faster to concatenate the images along the batch dimension than to enco

Extract depth maps using MoGe model for a sequence of images. Args: rgb_paths (list): List of paths to RGB images output

Given a single rgb_ts and a sorted array of depth_stamps, find the depth timestamp within 'max_diff' that is closest to 'rgb_ts'. Returns

:param flow: Optical flow tensor. :return: RGB image normalized between 0 and 1.

flow_utils.flowMaxNorm(flow): return the maximum of the l2-norm of the given flow ---- PARAMETERS ---- flow: the flow --

flow_utils.flowToColor(flow): return a color code flow field, normalized based on the maximum l2-norm of the flow flow_utils.flowToColor(flow

Generate tuples of overlapping views for a given scene. generate_depth: generate depth images and camera parameters.

Generate images from a metadata file for reproducibility purposes.

extract 3D_model (glb file) from a reconstructed scene