HyperMorph: Amortized Hyperparameter Learning for Image Registration
Hyperparameter tuning in learning-based registration is a time-consuming process and typically involves training many separate models with various hyperparameter values, potentially leading to suboptimal results. To address this inefficiency, we introduce HyperMorph: amortized hyperparameter learning for image registration, a strategy that learns the effects of hyperparameters on deformation fields, facilitating fast hyperparameter evaluation at test-time.
Visit the tutorial for information on developing and training HyperMorph models or the related demo on interactive hyperparameter tuning. Code can be found in the VoxelMorph and neurite repositories.
HyperMorph is published in the Journal of Machine Learning for Biomedical Imaging. If this work is useful to you, please cite:
Learning the Effect of Registration Hyperparameters with HyperMorph
Andrew Hoopes, Malte Hoffmann, Douglas N. Greve, Bruce Fischl, John Guttag, Adrian V. Dalca
Journal of Machine Learning for Biomedical Imaging, 2022.
HyperMorph was originally presented at the 2021 international conference on Information Processing in Medical Imaging (IPMI):
Hypermorph: Amortized Hyperparameter Learning for Image Registration
Andrew Hoopes, Malte Hoffmann, Bruce Fischl, John Guttag, and Adrian Dalca
International Conference on Information Processing and Medical Imaging, 12729, pp. 3-17. 2021.