Compute the features for horizontal edges, given the edge graph and the level-0 'subedges' making up each edge. The features computed here ar

Searches for the last occurrence of `OnTheFlyInstanceGraph` among the `on_device_transform` of the dataset and modifies the graph construction

Number of available virtual or physical CPUs on this system, i.e. user/real as output by time(1) when called with an optimally scaling usersp

Compute orthonormal bases for a set of 3D vectors. The 1st base vector is the normalized input vector, while the 2nd and 3rd vectors are const

labels: List of Tensors, each shape = (N,), level-0 point to higher level cluster IDs e.g. [label_lvl0, label_lvl1, label_lvl2]

Compute the radius neighbors of clusters. Two clusters are considered neighbors if 2 of their points are distant of `gap` of less. The un

Move the NAG with all Data in it to CPU.

Convert a CSR matrix to its dense counterpart of a given shape.

Detach all tensors in the CSRData.

Returns the device on which the logits are stored, assuming all other output variables held by the object are also on the same device.

Convert point-level edges into superedges between clusters, based on point-to-cluster indexing 'super_index'. Optionally 'edge_attr' can be pa

Given a graph of point segments, compute the concatenation of points belonging to either source or target segments for each edge of the segmen

Same as torch.repeat_interleave but relies numpy on CPU. This saves a little bit of time when manipulating large tensors on CPU. ``` from

Calculate nearest point to all focal axes in poses.

Helps `self.from_list()` and `self.to_list()` identify which classes to use for batch collation and un-collation.

get the most common label from a group of labels (except for 0) assign a new label when there exist three or more major label in a seg

Copied from Plenoxels Continuous learning rate decay function. Adapted from JaxNeRF The returned rate is lr_init when step=0 and lr_fina

sim: List of Tensors, each shape like superpoint topk: int, number of top similar points to consider level: int, choose certa

Optionally recursive outlier search. The `xyz_query` and `xyz_search`. Search for points with less than `k_min` neighbors within a radius of `

r"""Returns :obj:`True`, if the debug mode is enabled.

Check if the graph is 'trimmed': same as coalescing with the additional constraint that (i, j) and (j, i) edges are duplicates. :param edge_i

Convert `obj` to nested lists.

Read an HDF5 file of group produced using `dense_to_csr_hdf5` and return the dense tensor. An optional idx can be passed to only read correspo

Load an image and cast to float32.

Normalize an array of floats in a unit-hypercube of shared scale. Typically useful for visualizing float features with colors

Compute the oracle predictions for instance and panoptic segmentation. This is a proxy for the highest achievable performance with the

read XYZ RGB for each vertex. Note: RGB values are in 0-255

Render the scene. Background tensor (bg_color) must be on GPU!

Save an image (probably RGB) in [0, 1] to disk as a uint8 PNG.

Save torch.Tensor to HDF5 file. :param d: dictionary of 2D torch.Tensors :param f: h5 file path of h5py.File or h5py.Group :param key: st

Helper for scatter_mean with weights

For each pair of segments indicated in edge_index, find the 2 closest points between the two segments. NB: this is an approximate, iterative

Scatter implementation for PCA. Returns eigenvalues and eigenvectors for each group in idx. If x has shape N1xD and idx covers indices in [0,

Semantic predictions on the level-1 superpoint. Final semantic segmentation predictions are the argmax of the first-level partition l

Interactive data visualization. :param input: Data or NAG object :param path: str Path to save the visualization into a sharable HTML

Compute indices to sample elements in a set of size `idx.shape`, based on which segment they belong to in `idx`. The sampling operation is ru

Return the Data list

labels1: (s1, lr-1) labels2: (s2, lr-1)

Render the scene. Background tensor (bg_color) must be on GPU!

Remove the homogeneous bottom row from [..., 4, 4] pose matrices.

Construct lookat view matrix.

Partition a point cloud based on a regular XY grid. Returns, for each point, the index of the grid cell it falls into. :param pos: Tensor

Using the top 7 prompt list to evaluate the mask of replica.

Initialize the pointers and values. Values are passed as args and stored in a list. They are expected to all have the same size and s

Local helper to compute multiple instance partitions from the same input data, based on diverse partition parameter settings.

Group data based on indices in cluster. The option ``mode`` controls how data gets aggregated within each cluster. By default, some special k

Search neighboring segments with points distant from `gap`or less. :param nag: NAG Hierarchical structure :param k_min: int, List

:param nag: :param i_level: :param k_min: :param k_max: :param gap: :param trim: :param cycles: :param chunk_size:

Populate `self._slice_dict` and `self._inc_dict` with inferred collation for missing keys. Unlike PyG, we want to handle attributes t

Partition an instance graph using cut-pursuit. :param node_x: Tensor of shape [num_nodes, num_dim] Node features :param node_logits:

Compute all edges and edge features for a horizontal graph, given a trimmed graph and some precomputed edge attributes.

Compute edge features for a vertical graph, given child and parent nodes.