2D Gaussian Splatting

A radiance-field rendering method that represents a 3D scene as a collection of oriented, elliptical 2D Gaussian "surfels" rather than the 3D Gaussian blobs of the original Gaussian Splatting, trading a small amount of rendering flexibility for materially better surface-reconstruction accuracy while keeping fast, differentiable rasterization. Introduced by Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger & Shenghua Gao (arXiv:2403.17888), published as part of ACM Transactions on Graphics / SIGGRAPH 2024 — peer-reviewed, per the paper's own related-DOI record.

Matters to this vault as the primary baseline every promoted Hexels claim-note measures against: Hexels is pitched as a drop-in replacement for 2DGS's Gaussian weighting function, and its headline PSNR and Chamfer-distance gains are reported relative to 2DGS at matched memory budgets — while its headline F1 gain on Tanks and Temples is reported against unconstrained 2DGS, the baseline at its best setting rather than at a matched budget.

Updates

• 2026-10-30 — propagation-repair. The scoping of the headline Hexels-vs-2DGS gains was split. Was: "its headline PSNR, Chamfer-distance, and F1 gains are all reported relative to 2DGS at matched memory budgets". Now: PSNR and Chamfer-distance are the memory-matched comparisons, but the headline F1 gain (Hexels 34.38% vs 31.72%, Tanks and Temples) is against unconstrained 2DGS — memory-matched 2DGS scores 22.05% in the same Table 6. Propagates the 2026-10-30 cross-model correction to claim-hexels-surface-reconstruction-chamfer-distance-and-f1-gains-over-2dgs.

References

• claim-hexels-replace-gaussian-kernel-with-hexagonal-support-and-linear-color-field-in-2dgs
• claim-hexels-matched-memory-budget-psnr-gains-over-2dgs-and-triangle-splatting
• claim-hexels-surface-reconstruction-chamfer-distance-and-f1-gains-over-2dgs
• claim-hexels-2dgs-triangle-splatting-baselines-peer-reviewed-hexels-is-not