meshoptimizer updated for hierarchical clustered LOD to process billions of triangles

Background on NVIDIA's Zorah Demo

NVIDIA released RTX Mega Geometry with the Zorah demo, a 100 GB Unreal Engine scene demonstrating clustered raytracing and Nanite clustered LOD. This technology allows streaming and displaying highly detailed scenes with full raytracing without relying on Nanite proxy meshes for raytracing.

NVIDIA later released an update to their open-source vk_lod_clusters sample, which included the Zorah scene as a glTF file. This provided a more accessible format for developers to experiment with the underlying technology.

meshoptimizer Enhancements

The release of the Zorah scene in glTF format prompted improvements to meshoptimizer, specifically enhancing its support for hierarchical clustered LOD. This update enables the tool to efficiently handle extremely large 3D models, processing billions of triangles within minutes.

The core task involves generating a hierarchical structure for meshes, streaming parts of this structure at appropriate detail, and rendering visible sections at suitable detail levels. This approach scales to large models by distributing detail across multiple regions.

Hierarchical Clustered LOD Explained

Hierarchical clustered LOD uses a graph of clusters, where each cluster is a small set of triangles (up to 128) representing a mesh patch at a specific detail level. This structure supports multiple levels of detail across different mesh regions, ensuring efficient rendering.

The runtime system streams and renders these clusters, replacing a detailed cluster with a coarser one only if the visual error remains below one pixel. Temporal anti-aliasing (TAA) or other temporal filters hide these transitions.

Technical Challenges

The technique presents three main challenges: generating the hierarchical structure from a highly detailed mesh, compressing the resulting data for efficient storage and transmission, and managing the runtime streaming and rendering to maintain visual quality and performance.