Neural Shadow Mapping

TL;DR

This paper introduces a neural shadow mapping technique that produces high-quality hard and soft shadows efficiently, outperforming traditional methods and supporting dynamic scenes without additional post-processing.

Contribution

It presents a novel neural shadow mapping approach that is fast, memory-efficient, and robust to scene changes, eliminating the need for post-process anti-aliasing or denoising.

Findings

Achieves visual quality comparable to ray tracing.

Operates efficiently with memory bandwidth-aware architecture.

Supports dynamic scenes with changing geometry and lighting.

Abstract

We present a neural extension of basic shadow mapping for fast, high quality hard and soft shadows. We compare favorably to fast pre-filtering shadow mapping, all while producing visual results on par with ray traced hard and soft shadows. We show that combining memory bandwidth-aware architecture specialization and careful temporal-window training leads to a fast, compact and easy-to-train neural shadowing method. Our technique is memory bandwidth conscious, eliminates the need for post-process temporal anti-aliasing or denoising, and supports scenes with dynamic view, emitters and geometry while remaining robust to unseen objects.