/Nonrigid-ICP-Pytorch

Non-rigid alignment of two depth scans using Adam

Primary LanguageC++MIT LicenseMIT

Requirements

  • Pytorch (tested on 1.9.0)
  • Pytorch3D
  • lietorch
  • Mayavi & open3D for visualization

install

cd cxx
python setup.py install

Run on the moose example, leveraging the sparse landmark correspondences predicted by Leaprd

python main.py config.yaml

Config the w_ldmk to 0 to get results of the vanilla N-ICP.

Point clouds before and after registration
drawing
drawing