Datasets

Normal Datasets

NTU([root, num_classes, eval_type, split, ...])	NTU Dataset
UCLA([root, split, transform])	NW-UCLA Dataset.

Functional Datasets

DiskCache(dataset[, root, transform])	Cache Dataset instance to disk.
Apply(dataset, transform)	Apply Transform to Dataset instance.

class torch_skeleton.datasets.Apply(dataset: Dataset, transform: Callable)

Apply Transform to Dataset instance.

Parameters

• dataset (Dataset) – dataset to apply transform to

• transform (Transform) – transform to apply

class torch_skeleton.datasets.BABEL(root: str = '.', num_classes: int = 60, split: str = 'train', extra: bool = False, transform: Optional[Callable] = None)

BABEL Dataset

Downloads pre-processed datasets

Parameters

• root (str) – root directory of dataset

• num_classes (int) – number of classes, 60 for BABEL60, 120 for BABEL120

• extra (bool) – flag to use extra data

• split (str) – split type. either "train" or "val" or "test"

• transform (Transform) – transform to apply to dataset

class torch_skeleton.datasets.DiskCache(dataset: Dataset, root: str = '.', transform: Optional[Callable] = None)

Cache Dataset instance to disk.

Caches output of dataset to disk by creating a temporary directory at root.

Parameters

• root (str) – root directory of cache

• dataset (Dataset) – dataset to cache

class torch_skeleton.datasets.NTU(root: str = '.', num_classes: int = 60, eval_type: str = 'camera', split: str = 'train', transform: Optional[Callable] = None)

NTU Dataset

Parameters

• root (str) – root directory of dataset

• num_classes (int) – number of classes, 60 for NTU60, 120 for NTU120

• eval_type (str) – evaluation type for train/val split. "subject" for cross-subject, "camera" for cross-view.

• split (str) – split type. either "train" or "val"

• transform (Transform) – transform to apply to dataset

class torch_skeleton.datasets.UCLA(root='.', split='train', transform: Optional[Callable] = None)

NW-UCLA Dataset.

Parameters

• root (str) – root directory of dataset

• split (str) – split type, either "train" or "val"

• transform (Transform) – transform to apply to dataset