Neurosymbolic Models for Computer Graphics

Procedural models (i.e. symbolic programs that output visual data) are a historically-popular method for representing graphics content: vegetation, buildings, textures, etc. They offer many advantages: interpretable design parameters, stochastic variations, high-quality outputs, compact representation, and more. But they also have some limitations, such as the difficulty of authoring procedural model from scratch. More recently, AI-based methods, especially neural networks, become popular creating graphic content. These techniques allow users to directly specify desired properties artifact want create (via examples, constraints, or objectives), while search, optimization, learning algorithm takes care details. However, this ease use comes at cost, it's often hard interpret manipulate these representations. In state-of-the-art report, we summarize research on neurosymbolic in computer graphics: methods combine strengths both AI represent, generate, data. We survey recent work applying represent 2D shapes, 3D materials & textures. Along way, situate each prior unified space models, which helps reveal underexplored areas opportunities future research.