vis4d.data

The data package defines the full data pipeline.

We provide dataset implementations in the datasets submodule that return a common data format DictData. This data format is used by the pre-processing functions in the submodule transforms. The preprocessing functions are composed with the datasets in DataPipe. Optionally, a reference view sampler can be added here. The DataPipe is input to torch.data.DataLoader, for which we provide utility functions for instantiation that handle also batch-wise preprocessing and batch collation.

Modules

vis4d.data.cbgs	Class-balanced Grouping and Sampling for 3D Object Detection.
vis4d.data.const	Defines data related constants.
vis4d.data.data_pipe	DataPipe wraps datasets to share the prepossessing pipeline.
vis4d.data.datasets	Datasets module.
vis4d.data.io	Init io module.
vis4d.data.iterable	Iterable datasets.
vis4d.data.loader	Dataloader utility functions.
vis4d.data.reference	Reference View Sampling.
vis4d.data.resample	Resample index to recover the original dataset length.
vis4d.data.samplers	Vis4D data samplers.
vis4d.data.transforms	Transforms.
vis4d.data.typing	Type definitions related to the data pipeline.