search

facebookresearch / pytorch3d

PyTorch3D is FAIR's library of reusable components for deep learning with 3D data
https://pytorch3d.org/
Other
8.81k stars 1.32k forks source link

Request to Add DirDist to pyTorch3D #1739

Open rsy6318 opened 8 months ago

rsy6318 commented 8 months ago

We have proposed a new distance metric for 3D geometry data (point clouds and triangle meshes), DirDist. Different from previous metrics, our DirDist concentrates on the implicit fields of the given shapes. We have combine DirDist with various applications, including Template Surface Fitting, Rigid Registration, Non-rigid Registration, Scene Flow Estimation, and Human pose optimization, and get better results than previous distance metrics. See more details in our paper [Arxiv].

As a general distance metric, we think that our DirDist could boost the development of 3D geometry modeling, and we hope that our DirDist could be included in pytorch3d library.

rsy6318 commented 8 months ago

This is just an initial version, and there will be updates later to make the code style consistent with PyTorch3D.

rsy6318 commented 8 months ago

I have added our DirDist to my forked pytorch3D. Our DirDist outperform various distance metrics, such as CD, EMD, on various 3D tasks, and the experiment results could be found in our manuscripts. We believe that our DirDist could boost the field of 3D modeling.