huangjh-pub / synorim

[T-PAMI'22] [Jittor & Pytorch] Multiway Non-rigid Point Cloud Registration via Learned Functional Map Synchronization
https://cg.cs.tsinghua.edu.cn/people/~huangjh/publication/synorim/
118 stars 10 forks source link

Visualization Code #2

Closed vhvkhoa closed 2 years ago

vhvkhoa commented 2 years ago

Thank you very much for your high impact work. The results are very impressive. I would like to ask for the visualization code or how to visualize input point clouds and the registered point cloud at output ?

tolgabirdal commented 2 years ago

We visualize the point clouds using a custom Blender script, which we have not released. If MeshLab is not enough for you, I would direct you to: https://github.com/tolgabirdal/Mitsuba2PointCloudRenderer. This has similar render quality.

Maybe @heiwang1997 also has something to add here.

vhvkhoa commented 2 years ago

Thank you so much for the clarification. But I think the current source code only supports training and evaluating on pre-processed dataset in .npz format.

Can I ask you for some instruction to do inference with SyNoRiM on the input point clouds to generate registered point cloud, then view inputs and output ?

heiwang1997 commented 2 years ago

Hi @vhvkhoa, thanks for your interest in our work. I have just pushed a modification to the evaluation script. If you run it with an additional --visualize flag, you will then be able to view all pairwise scene flows inferred by our method. The visualization is based on open3d -- you can install it via pip.

Thanks.