otx.data.utils#

Utility modules for core data modules.

Functions

adapt_tile_config(tile_config, dataset, task)

Config tile parameters.

adapt_input_size_to_dataset(dataset[, task, ...])

Compute appropriate model input size w.r.t.
otx.data.utils.adapt_input_size_to_dataset(dataset: Dataset, task: OTXTaskType = OTXTaskType.DETECTION, input_size_multiplier: int | None = None) → tuple[int, int][source]#

Compute appropriate model input size w.r.t. dataset statistics.

Parameters:

  • dataset (Dataset) – Datumaro dataset including all subsets.

  • task (OTXTaskType, optional) – Task type of the model. Defaults to OTXTaskType.DETECTION.

  • downscale_only (bool, optional) – Whether to allow only smaller size than default setting. Defaults to True.

  • input_size_multiplier (int | None, optional) – Multiplier for input size. If it’s set, return the input size which can be divisible by the value. Defaults to None.

Returns:

Recommended input size based on dataset statistics.

Return type:

tuple[int, int]

otx.data.utils.adapt_tile_config(tile_config: TileConfig, dataset: Dataset, task: OTXTaskType) → None[source]#

Config tile parameters.

Adapt based on annotation statistics. i.e. tile size, tile overlap, ratio and max objects per sample

Parameters:

  • tile_config (TileConfig) – tiling parameters of the model

  • dataset (Dataset) – Datumaro dataset including all subsets

  • task (Task) – task type of the model