An Inverse Procedural Modeling Pipeline for SVBRDF Maps

Procedural modeling is now the de facto standard of material in industry. models can be edited and are easily extended, unlike pixel-based representations captured materials. In this paper, we present a semi-automatic pipeline for general proceduralization. Given Spatially-Varying Bidirectional Reflectance Distribution Functions (SVBRDFs) represented as sets pixel maps, our decomposes them into tree sub-materials whose spatial distributions encoded by their associated mask maps. This decomposition maps progresses hierarchically, driven new spectrum-aware matting instance-based methods. Each decomposed sub-material proceduralized novel multi-layer noise model to capture local variations at different scales. Spatial these modeled either by-example inverse synthesis method recovering Point Process Texture Basis (PPTBF) or via random sampling. To reconstruct procedural propose differentiable rendering-based optimization that recomposes all generated procedures together maximize similarity between input We evaluate on variety synthetic real demonstrate method's capacity process wide range types, eliminating need artist designed graphs required previous work. As fully models, results expand arbitrary resolution enable high level user control appearance.