TransGI: Real-Time Dynamic Global Illumination With Object-Centric Neural Transfer Model

TL;DR

TransGI introduces a real-time neural rendering method that combines an object-centric neural transfer model with an efficient lighting system, enabling high-fidelity global illumination under dynamic lighting with low latency.

Contribution

It presents a novel neural rendering approach that supports glossy effects and dynamic lighting in real-time, overcoming limitations of previous methods in expressiveness and efficiency.

Findings

Achieves less than 10 ms per frame rendering time.

Provides high-fidelity global illumination with glossy effects.

Outperforms baseline methods in rendering quality.

Abstract

Neural rendering algorithms have revolutionized computer graphics, yet their impact on real-time rendering under arbitrary lighting conditions remains limited due to strict latency constraints in practical applications. The key challenge lies in formulating a compact yet expressive material representation. To address this, we propose TransGI, a novel neural rendering method for real-time, high-fidelity global illumination. It comprises an object-centric neural transfer model for material representation and a radiance-sharing lighting system for efficient illumination. Traditional BSDF representations and spatial neural material representations lack expressiveness, requiring thousands of ray evaluations to converge to noise-free colors. Conversely, real-time methods trade quality for efficiency by supporting only diffuse materials. In contrast, our object-centric neural transfer model achieves compactness and expressiveness through an MLP-based decoder and vertex-attached latent features, supporting glossy effects with low memory overhead. For dynamic, varying lighting conditions, we introduce local light probes capturing scene radiance, coupled with an across-probe radiance-sharing strategy for efficient probe generation. We implemented our method in a real-time rendering engine, combining compute shaders and CUDA-based neural networks. Experimental results demonstrate that our method achieves real-time performance of less than 10 ms to render a frame and significantly improved rendering quality compared to baseline methods.