Returns the dim'th component of the idx'th point in the class:

r"""Rigid transform to points and normals (optional). Given a point cloud P(3, N), normals V(3, N) and a transform matrix T in the form of

Helper accessor to the dataset points:

r"""Advanced index select. Returns a tensor `output` which indexes the `data` tensor along dimension `dim` using the entries in `index` which

Must return the number of data points

r"""Initialize a group normalization block. Args: num_groups: number of groups num_channels: feature dimension

Pools features from the closest neighbors. WARNING: this function assumes the neighbors are ordered. Args: x: [n1, d] features matri

* Builds the index */

r"""Move all tensors to cuda.

Select all the corners for each anchor point. A corner should be formed by the anchor point and its two nearby points. Input: -

Frees the previously-built index. Automatically called within * buildIndex(). */

r"""Decompose transformation matrix into rotation matrix and translation vector. Args: transform (Tensor): (*, 4, 4) Returns:

r"""Pairwise distance of two (batched) point clouds. Args: x (Tensor): (*, N, C) or (*, C, N) y (Tensor): (*, M, C) or (*, C, M)

r"""Computes neighbors for a batch of q_points and s_points, apply radius search (in stack mode). This function is implemented on CPU. Args:

r"""Compute Skew-symmetric Matrix. [v]_{\times} = 0 -z y z 0 -x -y x 0 Args: inputs (Ten

* Called during search to add an element matching the criteria. * @return true if the search should be continued, false if the results are * suf

Input, - idxs: (N,), instance index - instances: (X,), instance label of each point

r""" Compute matching matrix and score matrix for each patch correspondence. Input: - score_mat (Tensor): (B, K, K), or (B, K+1, K+1) if i

positive are 1. others are all zero.

instance_names: [0,3,5,...max_instance_idx], the index is not continuous idxs: [0,1,2,3...num_instances-1], the index is continuous

Create all edges for the graph topology. Input: - feats: (N1+N2+...+NB, in_channels) - batches: (B+1)

* Create a tree node that subdivides the list of vecs from vind[first] * to vind[last]. The routine is called recursively on each sublist. *

Loads a previous index from a binary file. * IMPORTANT NOTE: The set of data points is NOT stored in the file, so the * index object must be c

graph: {'nodes':{idx:Instance},'edges':{idx:idx}}

Input, - pred_nodes: (M,3),[scan_id,src_idx,tar_idx] Output dict, - output[matching_ptr]: (M,), matching ptr

Max pooling from neighbors. Args: x: [n1, d] features matrix neighbor_indices: [n2, max_num] pooling indices Returns:

r"""Prepare hierarchy points and features for the scene graph. It is a module built in GeoTransformer. It downsample the points and features i

Compute outliers using the estimated transform. Update correspondence scores by removing outliers.

Input, - batch_points_s: (M,K,3), source points - batch_points_t: (M,K,3), target points - batch_match_ptr: (M,), bat

Input, - scan_x_src: (N,D) - K: int

Input, - pos_feats1: (N, D), torch.Tensor - pos_feats2: (N, D), torch.Tensor - neg_feats: (M, D), torch.

Sample K points for each instance. If the number of points is less than K, fill the rest with invalid value. Input: - points_ins

Stores the index in a binary file. * IMPORTANT NOTE: The set of data points is NOT stored in the file, so when * loading the index object it m