Triangle Splatting+: Differentiable Rendering with Opaque Triangles

Triangle Splatting+ redefines 3D scene reconstruction and rendering by directly optimizing opaque triangles, the fundamental primitive of computer graphic, in a fully differentiable framework. Unlike Gaussian Splatting or NeRF-based approaches, it delivers real-time, game-engine-ready meshes without post-processing, enabling instant compatibility with engines like Unity or Unreal.

Developed by researchers from the University of Liège, Simon Fraser University, University of Maryland, University of British Columbia, University of Toronto, and Adobe Research, this method bridges radiance-field optimization and traditional graphics pipelines.

Key Highlights:

Opaque Triangles, Differentiably Optimized: Achieves state-of-the-art mesh-based novel view synthesis with only opaque primitives, eliminating the transparency issues of prior splatting methods.
Shared-Vertex Connectivity: Introduces a vertex-based parametrization allowing triangles to share vertices and improving mesh structure with visual coherence.
No Post-Processing Needed: Outputs colored, connected, and simulation-ready meshes that integrate directly into game engines.
High Performance: Trains up to 2× faster than MiLo while reaching 4–10 dB higher PSNR on Mip-NeRF360 and Tanks & Temples datasets.

Why It Matters:

Triangle Splatting+ unites radiance fields and graphics primitives, enabling seamless integration of neural rendering into VR, AR, and game-engine pipelines. By turning triangles into differentiable, trainable building blocks, it opens the door to interactive 3D environments, physics-based simulations, and real-time digital twins that are both accurate and efficient.

Explore More:

Paper: arXiv:2509.25122

Project Page: https://trianglesplatting2.github.io/trianglesplatting2/