niftynet.layer.rand_elastic_deform module

Data augmentation using elastic deformations as used by: Milletari,F., Navab, N., & Ahmadi, S. A. (2016) V-net: Fully convolutional neural networks for volumetric medical image segmentation

class RandomElasticDeformationLayer(num_controlpoints=4, std_deformation_sigma=15, proportion_to_augment=0.5, spatial_rank=3)

Bases: niftynet.layer.base_layer.RandomisedLayer

generate randomised elastic deformations along each dim for data augmentation

__init__(num_controlpoints=4, std_deformation_sigma=15, proportion_to_augment=0.5, spatial_rank=3)

This layer elastically deforms the inputs, for data-augmentation purposes.

Parameters:

• num_controlpoints –
• std_deformation_sigma –
• proportion_to_augment – what fraction of the images to do augmentation on
• name – name for tensorflow graph

(may be computationally expensive).

randomise(image_dict)

layer_op(inputs, interp_orders, *args, **kwargs)