A Hierarchical Architecture for Neural Materials

TL;DR

This paper presents a hierarchical neural appearance model that improves the accuracy of reproducing complex material appearances, especially with shadowing and highlights, using multi-scale features and frequency encoding.

Contribution

The proposed model introduces an inception-based core network with frequency encoding and gradient-based loss for enhanced neural material rendering accuracy.

Findings

Improved rendering of materials with shadowing and highlights.

Effective multi-scale feature capture through inception modules.

Demonstrated success on synthetic and real-world examples.

Abstract

Neural reflectance models are capable of reproducing the spatially-varying appearance of many real-world materials at different scales. Unfortunately, existing techniques such as NeuMIP have difficulties handling materials with strong shadowing effects or detailed specular highlights. In this paper, we introduce a neural appearance model that offers a new level of accuracy. Central to our model is an inception-based core network structure that captures material appearances at multiple scales using parallel-operating kernels and ensures multi-stage features through specialized convolution layers. Furthermore, we encode the inputs into frequency space, introduce a gradient-based loss, and employ it adaptive to the progress of the learning phase. We demonstrate the effectiveness of our method using a variety of synthetic and real examples.