Dataset returns a point cloud from a train or test split from one sequence from a raw Mulran dataset

Multiple NCLT sequences indexed as a single dataset. Each element is identified by a unique global index.

Dataset returns a point cloud from a train or test split from one sequence from a raw Mulran dataset

Multiple Oxford sequences indexed as a single dataset. Each element is identified by a unique global index.

Randomly remove part of the point cloud. Similar to PyTorch RandomErasing but operating on 3D point clouds. Erases fronto-parallel cuboid.

Provides intrinsic parameters and undistortion LUT for a camera. Attributes: camera (str): Name of the camera. camera sensor (str

Provides intrinsic parameters and undistortion LUT for a camera. Attributes: camera (str): Name of the camera. camera sensor (str

Dimensionality reduction as a convolutional layer :param int input_dim: Network out_channels :param in dim: Whitening out_channels, for dimen

Compute the attention as L2-norm of local descriptors

An attention module used in BEVFormer based on Deformable-Detr. `Deformable DETR: Deformable Transformers for End-to-End Object Detection. <ht

NetVLAD layer implementation

Average pooling that smoothens the feature map, keeping its size :param int kernel_size: Kernel size of given pooling (e.g. 3)

Parameters for model training