Home

Awesome

Non-rigid Point Cloud Registration with Neural Deformation Pyramid [Paper]

Hierarchical non-rigid registration of multiple scans

<img src="img/dino.gif" alt="drawing" width="650"/>

Scale variant non-rigid registration with Sim(3) warp field

<img src="img/transfer.gif" alt="drawing" width="800"/>

Requirements

The code tested on python=3.8.10, pytorch=1.9.0 with the following packages:

Obtain the 4DMatch benchmark

ln -s /path/to/4Dmatch  ./data

Reproduce the result of NDP (no-learned)

python eval_nolearned.py --config config/NDP.yaml  

To visualize the registration result, add --visualize.

Reproduce the result of LNDP (supervised)

cd correspondence/cpp_wrappers; sh compile_wrappers.sh; cd ../..
python eval_supervised.py --config config/LNDP.yaml  

To visualize the registration result, add --visualize.

Run shape transfer example

python shape_transfer.py -s sim3_demo/AlienSoldier.ply -t sim3_demo/Ortiz.ply

Our related projects

Lepard: rabbityl/lepard
DeformingThings4D: rabbityl/DeformingThings4D

Citation

If you use our code please cite:

@article{li2022DeformationPyramid, 
    title={Non-rigid Point Cloud Registration with Neural Deformation Pyramid.}, 
    author={Yang Li and Tatsuya Harada},
    journal={arXiv preprint arXiv:2205.12796},
    year={2022}
}