3D Scene Generation from Scene Graphs and Self-Attention

A novel attention-based conditional variational autoencoder (cVAE) model that generates diverse and plausible 3D scene layouts from input scene graphs.