HyperMorph

Overview

HyperMorph is the hypernetwork-conditioned VoxelMorph: the regularization weight is an input of the network, so the smoothness of the deformation is chosen at inference time. MIR’s HyperVxmDense is the convolutional counterpart of HyperTransMorph.

Usage

from MIR.models import HyperVxmDense
import MIR.models.configs_VoxelMorph as configs_VoxelMorph

config = configs_VoxelMorph.get_VXM_default_config()
config.img_size = fix.shape[2:]
model = HyperVxmDense(config).cuda()
flow = model((mov, fix), torch.tensor([[0.3]], device='cuda'))   # regularization weight

Key modules

• MIR.models.HyperMorph, MIR.models.configs_VoxelMorph

Related tutorials

• IXI HyperMorph

References

• Hoopes, Hoffmann, Fischl, Guttag, Dalca. HyperMorph: amortized hyperparameter learning for image registration. IPMI 2021.