Learning-Based Inverse Bi-Scale Material Fitting From Tabular BRDFs.

Area of Science:

• Computer Graphics
• Material Science
• Computational Imaging

Background:

• Material appearance design links small-scale structures (meso-scale geometry, micro-scale Bidirectional Reflectance Distribution Functions - BRDFs) to large-scale appearance (macro-scale BRDF).
• Manual adjustment of these small-scale elements for desired macro-scale BRDFs is often a time-consuming trial-and-error process.

Purpose of the Study:

• To develop an efficient, learning-based solution for fitting a target macro-scale BRDF using a combination of meso-scale geometry and micro-scale BRDFs.
• To address challenges in representing both large-scale macro-BRDFs (compactness, expressiveness) and small-scale details (diverse geometries, spatially varying micro-BRDFs).

Main Methods:

• Proposed a novel 2D subset of tabular BRDF representation for compact and expressive macro-scale BRDFs.
• Represented small-scale geometries and BRDFs in categorized, physically parameterized search spaces.
• Developed an end-to-end model for classification and parameter estimation, mapping macro-BRDFs to small-scale details.

Main Results:

• The learning-based solution achieves higher reconstruction accuracy compared to traditional fitting methods.
• The proposed method demonstrates a wider coverage of material appearance gamuts.
• Successfully reconstructs target macro-scale BRDFs from combined meso-scale geometry and micro-scale BRDF inputs.

Conclusions:

• The end-to-end learning approach effectively bridges the gap between small-scale material properties and large-scale appearance.
• This method offers a significant improvement over manual or less integrated fitting techniques in material appearance design.
• Enables more efficient and accurate creation of materials with specific visual characteristics.