dataeval.utils.data.collate¶

dataeval.utils.data.collate(dataset: ImageClassificationDataset[TArray] | ObjectDetectionDataset[TArray], model: torch.nn.Module, device: torch.device | str | None = None, batch_size: int = 64) → tuple[torch.Tensor, Targets, list[dict[str, Any]]]¶

dataeval.utils.data.collate(dataset: ImageClassificationDataset[TArray] | ObjectDetectionDataset[TArray], model: None = None, device: torch.device | str | None = None, batch_size: int = 64) → tuple[list[TArray], Targets, list[dict[str, Any]]]

Collates a dataset to images/embeddings, targets and metadata.

Parameters:¶

dataset : ImageClassificationDataset or ObjectDetectionDataset¶

A dataset conforming to MAITE dataset protocols.

model : torch.nn.Module or None, default None¶

A torch model to use for encoding. If an encode() function is present on the model it will be called, otherwise it will use the __call__() function.

device : torch.device, str or None, default None¶

Device to use when encoding with the provided model.

batch_size : int, default 64¶

Batch sizes to use when encoding with the provided model.

Returns:¶

• Images as a list of original source data or embeddings as a torch.Tensor if encoded.

• Targets including labels, scores as well as boxes and source indices for objects.

• Metadata aggregated as a list of individual datum metadata dictionaries.

Return type:¶

tuple[torch.Tensor | list[Any], Targets, list[dict[str, Any]]]

Note

For more on supported image classification and object detection dataset protocols, see protocol documentations for MAITE.