Taken from https://github.com/colmap/colmap/blob/dev/scripts/python/read_dense.py :param path: path to the colmap binary file. :return:

Recenter poses around the origin.

Returns a copy of self with modified pointers to account for new groups. Affects the num_groups and the order of groups. Injects 0-len

Return a new InstanceData with void clusters, objects and pairs removed. IMPORTANT: By convention, we assume `y ∈ [0, num_cla

Run a command-line process from Python and print its outputs in an online fashion. Credit: https://www.endpointdev.com/blog/2015/01/getting-r

Return the predicted and target edge affinities, along with masks indicating same-class and same-object edges. The output is sanitized

Save Batch to HDF5 file. :param f: h5 file path of h5py.File or h5py.Group :param y_to_csr: bool Convert 'y' to CSR forma

Save CSRBatch to HDF5 file. :param f: h5 file path of h5py.File or h5py.Group :param fp_dtype: torch dtype Data type to w

Returns a new Cluster with updated clusters and points, which indexes `self` using entries in `idx`. Supports torch and numpy fancy in

Returns a new Data with updated clusters, which indexes `self` using entries in `idx`. Supports torch and numpy fancy indexing. `idx`

Indexing CSRBatch format. Supports Numpy and torch indexing mechanisms. Since indexing breaks batching, this will return a CSRData

Compute the oracle performance for semantic segmentation, when all clusters predict the dominant label among their points. This corres

Semantic target on the level-1 superpoint. Final semantic segmentation target are the label histogram of the first-level partition lo

This assumes all items in `self.obj_index_pred` follow the output format of `InstancePartitioner._grid_forward()`.

See `src.visualization.show`.

Returns the size of each group (i.e. the pointer jumps).

By default, starmap only accepts args and not kwargs. This is a helper to get around this problem. :param fn: callable The function t

Cluster object indicating subpoint indices for each point.

Index of the superpoint each point belongs to.

Panoptic predictions on the level-1 nodes. Returns the predicted semantic label and instance index for each superpoint, along with the

Optional decorator that wraps the task function in extra utilities. Makes multirun more resistant to failure. Utilities: - Calling the `

Move the CSRBatch to the specified device.

:param labels0: (s1, lr-1), label distribuion of one seg :param labels1: (s2, lr-1), label distribution of neighbors of the seg above :r

:param seg_seen: (S, M), ratio of seen part of primitives in every view :param seg_neighbors: (S, S), 1 if two segs are neighbors :param

Decorator intended for numpy-based functions to be fed and return torch.Tensor arguments. :param func: :return:

project N (1e6) points to M (1e3) images :param points_world: (N, 3) :param color_intrinsic, depth_intrinsic: (M, 3, 3) the intrinsics of col

Learning rate scheduling per step

Vertical edge features.

All keys starting with `v_edge_`.

List of names for each value. These will be used as labels for serialization.

Returns a mask on the edges indicating those connecting two void nodes.

Returns a mask on the level-1 nodes indicating which is void. By convention, nodes/superpoints are void if they contain more than 50%

Semantic predictions on the level-0 voxels. Final semantic segmentation predictions are the argmax of the first-level partition logit

project N points to M images :param points_world: (N, 3) :param color_intrinsic, depth_intrinsic: (M, 3, 3) the intrinsics of color and depth