For non-Euclidean data such as meshes of humans, a prominent task for generative models is geometric disentanglement; the separation of latent codes for intrinsic (i.e. identity) and extrinsic (i.e. pose) geometry. This work introduces a novel mesh feature, the conformal factor and normal feature (CFAN), for use in mesh convolutional autoencoders. We further propose CFAN-VAE, a novel architecture that disentangles identity and pose using the CFAN feature and parallel transport convolution. CFAN-VAE achieves this geometric disentanglement in an unsupervised way, as it does not require label information on the identity or pose during training. Our comprehensive experiments, including reconstruction, interpolation, generation, and canonical correlation analysis, validate the effectiveness of the unsupervised geometric disentanglement. We also successfully detect and recover geometric disentanglement in mesh convolutional autoencoders that encode xyz-coordinates directly by registering its latent space to that of CFAN-VAE.
View on arXiv