Dataset

from kagu.datasets.kagu_dataset import KaguDatasetArguments, KaguDataset

kagu_args = KaguDatasetArguments(dataset='data/kagu',
                                  frame_size=[299,299],
                                  world_size=4,
                                  global_rank=0,
                                  snippet=16,
                                  step=8)
kagu_dataset = kagu_dataset(kagu_args)

class kagu.datasets.kagu_dataset.KaguDatasetArguments(dataset: str, frame_size: list, world_size: int, global_rank: int, snippet: int, step: int)

Arguments for Kagu dataset

dataset: str

The path to the datset

frame_size: list

The frame size of the images

world_size: int

The number of proceses spawned

global_rank: int

The rank of the current process

snippet: int

Number of frames to process

step: int

The stride by which we process the frames. Same as snippet if not overlapping

class kagu.datasets.kagu_dataset.KaguDataset(args: KaguDatasetArguments)

The kagu dataset class that streams video files and outputs a list of frames according to snippet and step.

Parameters:

args (KaguDatasetArguments) – The parameters used for the Kagu dataset

__getitem__(index)

Iterates over the dataset in a streaming fashion and retrieves KaguDatasetArguments.snippet frames at a time.

Parameters:

index (int) – The index of the item in the dataset to retrieve. Not used.

Returns:

• (torch.tensor): the frames in tensor format