vp_suite.utils.dataset_wrapper

class VPDatasetWrapper(dataset_class, split, **dataset_kwargs)

Bases: object

A class that wraps VPDataset instances to handle training and testing data in the same way.

ALLOWED_SPLITS = ['train', 'test']

On creation, the wrapper must be initialized with one of these split values.

property NAME

Returns the dataset name.

__init__(dataset_class, split, **dataset_kwargs)

Instantiates the dataset class specified by the given dataset_class value and split identifier.

Parameters

• dataset_class (Any) – A string identifier corresponding to the dataset class that should be instantiated.

• split (str) – A string specifying whether this Wrapper should wrap a training/validation or a test set.

• **dataset_kwargs (Any) – Additional optional dataset configuration arguments.

property action_size

Returns the dataset’s action size (which is 0 for datasets that don’t provide actions)

property config

Returns a dictionary containing the dataset’s configuration parameters.

property data_dir

Returns the dataset’s data location

property img_shape

Returns the shape of each frame of the sequences that will be provided by the dataset (after preprocessing)

is_test_set()

Returns True if this wrapper wraps testing fata, false otherwise.

is_training_set()

Returns True if this wrapper wraps training/validation data, false otherwise.

reset_rng()

Resets the RNG for all wrapped datasets.

set_seq_len(context_frames, pred_frames, seq_step)

Sets the desired sequence length for all wrapped datasets by calculating the needed sequence length from the provided sequence parameters

Parameters

• context_frames (int) – Number of context/input frames (these will be provided as input to the VPModel)

• pred_frames (int) – Number of frames the VPModel has to predict, given the context frames.

• seq_step (int) – Sequence step (assembling the sequence from every Nth frame or the original video)

property test_data

Returns the wrapped test dataset. If not existent, raises an error.

property train_data

Returns the wrapped training dataset. If not existent, raises an error.

property val_data

Returns the wrapped validation dataset. If not existent, raises an error.