Some issues about papaer

[CRISZJ]

Good job and thanks for your code. After reading, I have a question about the paper. Can we use VAE's encoder and decode to replace AE in the pipeline, that is, can we encode Partial shape into a distribution and instead of a Latent space?

looking forward to your reply. thanks again.

[ChrisWu1997]

I think that would hurt the performance, in terms of the final point cloud quality. The purpose of using a pretrained AE is to get a good latent representation for point clouds such that GAN can be trained on latent space effectively. VAE has worse reconstruction ability compared to AE, i.e. the decoded point clouds from VAE are of worse quality.