`nidl.volume.backbones`: Available backbones¶

This modules details the public API you should use and implement for a nidl compatible backbone, as well as the backbones available in nidl.

Introduction¶

An backbone is a torch.nn.Module class and contains definitions of models for addressing different tasks, including image classification, image segmentation, represnetation learning…

Pre-trained weights¶

nidl offers pre-trained weights for every provided architecture. The weights can be restored using the nidl.utils.weights.Weights.load_pretrained method. Available weights are listed here.

Weights¶

Classes that allow to restore pre-trained weights.

Weights(name, data_dir, filepath)

A class to handle (retrieve and apply) model weights or lightning checkpoints.

Volume¶

Classes that implement architectures that can be applied on brain volumes and various utility functions.

`AlexNet`([n_embedding, in_channels])	3D AlexNet architecture adapted from Abrol et al., 2021.
`DenseNet`([growth_rate, block_config, ...])	3D Densenet architecture adapted from Huang et al. 2018.
`ResNet`(block, layers[, in_channels, ...])	3D ResNet architecture adapted from He et al. 2015.
`ResNetTruncated`(*args[, depth])	3D truncated ResNet-18 architecture adapted from He et al. 2015.

        classDiagram
  Module <|-- AlexNet
  Module <|-- DenseNet
  Module <|-- ResNet
  ResNet <|-- ResNetTruncated

`densenet121`(**kwargs)	3D Densenet-121 model adapted from Huang et al. 2018.
`resnet18_trunc`(depth, **kwargs)	3D truncated ResNet-18 architecture adapted from He et al. 2015.
`resnet50`(**kwargs)	3D ResNet-50 architecture adapted from He et al. 2015.
`resnet50_trunc`(depth, **kwargs)	3D truncated ResNet-50 architecture adapted from He et al. 2015.

Surface¶

Classes that implement architectures that ccan be applied on brain surfaces and various utility functions.

coming soon