LDDMM with PyTorch and KeOps

Large deformation diffeomorphic metric mapping (LDDMM) in PyTorch with KeOps library.
More friendly wrapper based on this tutorial:
https://www.kernel-operations.io/keops/_auto_tutorials/surface_registration/plot_LDDMM_Surface.html

Installation

Navigate to main direction and run

pip install .

Now you can import lddmm and enjoy.

Alternatively, you can install this without cloning with:

pip install git+https://github.com/PatRyg99/lddmm

Getting started

import pyvista as pv
from lddmm.registrator import LDDMMRegistrator

source_mesh = pv.read("<source-mesh-path>")
target_mesh = pv.read("<target-mesh-path>")

# Rigid regsitration
source_mesh.points -= source_mesh.points.mean(axis=0)
target_mesh.points -= target_mesh.points.mean(axis=0)

# Define and run LDDMM registration
registrator = LDDMMRegistrator(source_mesh, target_mesh, sigma=20, device="cpu")
registrator.optimize(2)

Obtaining interpolated deformation timepoints:

# Get raw shooting results
shots = registrator.shoot(nt=15)

# Get shooting results and save as .vtp files
registrator.export_shoot("shots", nt=15)