Move the NAG with all Data in it to device.

Move the NAG with all Data in it to CPU.

Move the NAG with all Data in it to CUDA.

Detach all tensors in the NAG.

Return a new NAG instance containing the Data clones.

Load torch.Tensor from an HDF5 file. See `save_tensor` for writing such file. Options allow reading only part of the rows. :param f: h5 file

Indexing mechanism on the NAG. Returns a new copy of the indexed NAG, with updated clusters. Supports int, torch and numpy indexing.

Sanitize an iterable of string key into a sorted list of unique keys. This is necessary for consistently hashing key list arguments of some tr

Load NAG from an HDF5 file. See `NAG.save` for writing such file. Options allow reading only part of the data. NB: if relevant, a NAG

Compute the super_index linking the points at level 'low' with superpoints at level 'high'. Note: 'low=-1' is accepted when level-0 h

Read and unpack the next bytes from a binary file. :param fid: :param num_bytes: Sum of combination of {2, 4, 8}, e.g. 2, 6, 16, 30, etc.

Save torch.Tensor to HDF5 file. :param x: 2D torch.Tensor :param f: h5 file path of h5py.File or h5py.Group :param key: str h5py.

Convert an int, slice, list or numpy index to a torch.LongTensor.

Move the CSRData to the specified device.

Convert 1D torch.Tensor of int into plotly-friendly RGB format. This operation is deterministic on the int values.

Convert to 'trimmed' graph: same as coalescing with the additional constraint that (i, j) and (j, i) edges are duplicates. If edge attributes

Return device of first Data in NAG.

Return a 1D tensor of indices converting the CSR-formatted clustering structure in 'self' into the 'super_index' format.

Return the obj, count, and y of the majority instance in each cluster (i.e. the object with which it has the highest overlap). :param

Merge clusters based on `idx` and return the result in a new InstanceData object. :param idx: 1D torch.LongTensor or numpy.NDArray

Convert torch.Tensor to numpy while respecting some constraints on output dtype.

This is a TEMPORARY, HACKY method to be called wherever edge_keys may cause an issue.

Convert a tensor of sizes into the corresponding pointers. This is a trivial but often-required operation.

Convert np.ndarray to torch.Tensor.

Compute the number of points of level 'low' contained in each superpoint of level 'high'. Note: 'low=-1' is accepted when level-0 has

See `src.visualization.show`.

Evaluate spherical harmonics at unit directions using hardcoded SH polynomials. Works with torch/np/jnp. ... Can be 0 or more batch d

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

Checks whether a 1D tensor of indices contains duplicates.

Search k-NN of x_query inside x_search, within radius `r_max`. Optionally, passing `batch_search` and `batch_query` will ensure the neighbor s

Extend `torch_geometric.Data.to` to handle Cluster and InstanceData attributes.

Modifies the `model.partitioner` parameters with parameters passed in the `partition_kwargs` dictionary.

Vectorized equivalent of: >>> torch.cat([torch.arange(s, s + w) for w, s in zip(width, start)])

Convert features of the format M x N with N>=1 to an M x 3 tensor with values in [0, 1 for RGB visualization].

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

Helper function producing a boolean mask of size `y.shape[0]` indicating which of the `y` (labels if 1D or logits/probabilities if 2D) are amo

Checks whether a 1D tensor of indices contains dense indices. That is to say all values in [0, a.max] appear at least once in a.

Load Cluster from an HDF5 file. See `Cluster.save` for writing such file. Options allow reading only part of the clusters. Th

Load CSRData from an HDF5 file. See `CSRData.save` for writing such file. Options allow reading only part of the clusters. :p

Normalization helper function.

Pad [..., 3, 4] pose matrices with a homogeneous bottom row [0,0,0,1].

Print some info about a tensor. Used for debugging.

reconstruct radiance field given cameras

For the sake of visualization, we require that `NAGAddKeysTo` does not remove input `Data` attributes after moving them to `Data.x`, so we may

Search for nodes with no edges in the graph and connect them to their k nearest neighbors. Update self.edge_index and self.edge_attr a

view: view camera depth: depthmap

Same as torch.randperm, but relies on numba for CPU tensors. This may bring a x2 speedup on CPU for n >= 1e5. ``` from time import time

Compute indices to sample elements at `low`-level, based on which segment they belong to at `high`-level. The sampling operation is r

Helper to pass a boolean mask and an index, to make sure indexing using the mask will not entirely discard all elements of index.

Convert pre-sorted dense indices to CSR format.

Compute the Intersection over Union (IoU) and the individual size for each cluster-object pair in the data. This is typically needed f

Checks whether a 1D tensor of indices is sorted.

Return a boolean mask of size num_nodes indicating which node has no edge in edge_index.

Search k-NN for a 3D point cloud xyz. This search differs from `knn_2` in that it operates on a single cloud input (search and query are the s

Panoptic predictions on the level-1 superpoints. Return the predicted semantic score and label for each predicted instance, along wit

Convert torch.Tensor of float RGB values in [0, 1] to plotly-friendly RGB format. If alpha is provided, the output will be expressed in RGBA f

Lexicographic sort 1D src points based on index first and src values second. Credit: https://github.com/rusty1s/pytorch_scatter/issues/48

Convert to 'trimmed' graph: same as coalescing with the additional constraint that (i, j) and (j, i) edges are duplicates. If edge at

Compute the predicted semantic label, score and logits for each predicted instance. This involves computing, for each predicted instan

Extend the PyG.Data.__inc__ behavior on '*index*' and 'face' attributes to our 'super_index'. This is needed for maintaining clusters