[WIP] Start off by expressing my gratitude for the amazing pytorch3d library. However, while working with pytorch3d.ops.estimate_pointcloud_local_coord_frames function, I have noticed an issue with the description of the principal curvatures. More technically, the eigenvector corresponding to the minimal eigenvalue of covariance is exactly the normal vector, but neither the other two eigenvalues are principal curvatures. The current implementation does not accurately compute the curvature, which is a crucial metric in the field of 3D deep learning. Therefore, I have taken the liberty of developing codes for curvature algorithms based on pytorch3d for both meshes and point clouds. I believe that this algorithm would be a valuable addition to the library and would help fill the current gap in the available curvature computation methods.
[WIP] Start off by expressing my gratitude for the amazing pytorch3d library. However, while working with pytorch3d.ops.estimate_pointcloud_local_coord_frames function, I have noticed an issue with the description of the principal curvatures. More technically, the eigenvector corresponding to the minimal eigenvalue of covariance is exactly the normal vector, but neither the other two eigenvalues are principal curvatures. The current implementation does not accurately compute the curvature, which is a crucial metric in the field of 3D deep learning. Therefore, I have taken the liberty of developing codes for curvature algorithms based on pytorch3d for both meshes and point clouds. I believe that this algorithm would be a valuable addition to the library and would help fill the current gap in the available curvature computation methods.