I attempted to use D3D-VAE for shape reconstruction. Specifically, I extracted point clouds from 3D meshes and used the point encoder to encode them into latents, and then decoded them back into meshes. However, I couldn't get normal results. Therefore, I would like to know some details about the reconstruction part, like the details of the original mesh, point cloud sampling strategies(point number, whether they are normalized or not), and the training details of the VAE part. Thanks!
I attempted to use D3D-VAE for shape reconstruction. Specifically, I extracted point clouds from 3D meshes and used the point encoder to encode them into latents, and then decoded them back into meshes. However, I couldn't get normal results. Therefore, I would like to know some details about the reconstruction part, like the details of the original mesh, point cloud sampling strategies(point number, whether they are normalized or not), and the training details of the VAE part. Thanks!