utils.datasets.MTProphesee

utils.datasets.MTProphesee#

class MTProphesee(name: str, root='./data', batch_size=4, num_steps=128, time_step=16, num_load_file=8, num_workers=4)[source]#

Bases: PropheseeDataModule

Multi-target Prophesee’s gen1 and 1mpx datasets.

The packages are provided in a multi-target form, meaning that one example can contain target labels at multiple time steps. Records are split into fixed-step chunks that are returned sequentially. Intended for testing. There is single-target dataset for training.

Parameters:

  • name (str) – The name of the dataset to download. Supported gen1 and 1mpx.

  • root (str, optional) – The directory where datasets are stored. Defaults to “./data”.

  • batch_size (int, optional) – Number of elements in a batch. Defaults to 4.

  • num_steps (int, optional) – Number of frames. Defaults to 16.

  • time_step (int, optional) – Time between frames. Defaults to 16.

  • num_load_file (int, optional) – Number of concurrently open files in each thread. Defaults to 8.

  • num_workers (int, optional) – A positive integer will turn on multi-process data loading with the specified number of loader worker processes. Defaults to 4.

Raises:

ValueError – Invalid dataset name.

Methods

create_dataset

Initializes dataset
create_dataset(gt_files: List[str], data_files: List[str]) → IterableDataset[source]#

Initializes dataset

Parameters:

  • gt_files (List[str]) – List of files with targets

  • data_files (List[str]) – List of files with records

Returns:

Ready dataset

Return type:

IterableDataset