Encoder Decoder for a 3d mesh

[ck-amrahd]

Hi guys,
I want to train a model that can encode 3d mesh into latent space [fixed dimensions] and decode it to minimize the reconstruction error [similar to encoder-decoder architecture for 2d images]. I can use conv2d and TransposeConv2d for 2d images. I think I can use the GraphConv layer by transforming mesh into a graph but I am not sure how to map these variable numbers of vertices to a fixed dimensional latent space and not sure how to do the reverse [latent space to mesh] in the decoder. Any help, links, and suggestions will be helpful. Thank you.

[boqian-li]

hi, have u made any progress on this idea?