A real-time rendering method for high albedo anisotropic materials with multiple scattering

TL;DR

The paper tackles real-time rendering of participating media by learning to approximate the iterative integration of the radiative transfer equation. It decomposes inputs into density, transmission, and phase features, builds graded transmission fields with a 3D-CNN, and uses diffuse and highlight sampling templates to drive a backbone network that predicts center-point scattering distributions for thousands of template centers in parallel. A two-template backbone with an attention mechanism enables selective feature fusion across density, transmission, and phase channels, achieving interactive frame rates without sacrificing rendering quality. The approach demonstrates superior speed and quality relative to prior methods, with practical implications for real-time graphics involving clouds, smoke, skin, and other volumetric media.

Abstract

We propose a neural network-based real-time volume rendering method for realistic and efficient rendering of volumetric media. The traditional volume rendering method uses path tracing to solve the radiation transfer equation, which requires a huge amount of calculation and cannot achieve real-time rendering. Therefore, this paper uses neural networks to simulate the iterative integration process of solving the radiative transfer equation to speed up the volume rendering of volume media. Specifically, the paper first performs data processing on the volume medium to generate a variety of sampling features, including density features, transmittance features and phase features. The hierarchical transmittance fields are fed into a 3D-CNN network to compute more important transmittance features. Secondly, the diffuse reflection sampling template and the highlight sampling template are used to layer the three types of sampling features into the network. This method can pay more attention to light scattering, highlights and shadows, and then select important channel features through the attention module. Finally, the scattering distribution of the center points of all sampling templates is predicted through the backbone neural network. This method can achieve realistic volumetric media rendering effects and greatly increase the rendering speed while maintaining rendering quality, which is of great significance for real-time rendering applications. Experimental results indicate that our method outperforms previous methods.

Figures (6)

• Figure 1: Overall flow chart.
• Figure 2: Data processing module flow chart.
• Figure 3: 3D-CNN generates transmission fields.
• Figure 4: Light relationship of template point at position p.
• Figure 5: Template point volume influence.