3D Shape Variational Autoencoder Latent Disentanglement via Mini-Batch Feature Swapping for Bodies and Faces

TL;DR

This paper introduces a self-supervised 3D shape VAE training method that effectively disentangles identity features in faces and bodies, outperforming existing approaches in latent representation quality.

Contribution

The authors propose a novel mini-batch feature swapping technique for training 3D shape VAEs to achieve better disentanglement of identity features in faces and bodies.

Findings

State-of-the-art disentanglement of identity features achieved.

Method maintains high-quality shape reconstruction.

Outperforms existing disentanglement methods.

Abstract

Learning a disentangled, interpretable, and structured latent representation in 3D generative models of faces and bodies is still an open problem. The problem is particularly acute when control over identity features is required. In this paper, we propose an intuitive yet effective self-supervised approach to train a 3D shape variational autoencoder (VAE) which encourages a disentangled latent representation of identity features. Curating the mini-batch generation by swapping arbitrary features across different shapes allows to define a loss function leveraging known differences and similarities in the latent representations. Experimental results conducted on 3D meshes show that state-of-the-art methods for latent disentanglement are not able to disentangle identity features of faces and bodies. Our proposed method properly decouples the generation of such features while maintaining good representation and reconstruction capabilities.