loaders

The loaders config group determines the way each Modality is processed by some ModalityLoader to bring sample entries to tensor ready format in mml. More details on available loaders can be found in the API entry of modality_loaders. The config requires an equally named entry to the Modality that points to the respective ModalityLoader to be used for that Modality.

For most cases the default configuration should offer a sufficing solution, but specific file formats might require a different ModalityLoader. There is also some potential for efficiency improvements through testing different ModalityLoader s.

The following callbacks configuration files are currently available:

default

The default configuration determines the general fallback option for all modalities.

image
default: OpenCVImageLoader

  • determines how the image modality is loaded

  • tries to use open CV for image loading

mask
default: OpenCVMaskLoader

  • determines how masks are loaded

  • tries to use open CV for mask loading

class
default: ClassLoader

  • determines how the class modality is loaded

classes
default: MultiLabelClassLoader

  • determines how the classes modality is loaded

soft_classes
default: SoftLabelClassLoader

  • determines how the soft_classes modality is loaded

value
default: ValueLoader

  • determines how the value modality is loaded

numpy

image
default: NumpyArrayImageLoader

  • determines how the image modality is loaded

  • tries to use numpy for image loading

pillow

image
default: PillowImageLoader

  • determines how the image modality is loaded

  • tries to use Pillow for image loading

pillow_acc

image
default: AcceleratedPillowImageLoader

  • determines how the image modality is loaded

  • tries to use a modified Pillow routine for image loading

scikit

image
default: ScikitImageLoader

  • determines how the image modality is loaded

  • tries to use scikit for image loading

torch

image
default: PureTorchvisionImageLoader

  • determines how the image modality is loaded

  • tries to use torchvision for image loading

uni

image
default: CombinedModalityLoader