---
title: "At matched model-memory budgets, Hexels improve PSNR over 2D Gaussian Splatting by 4.53–7.38 dB and over Triangle Splatting by 1.70–2.90 dB across three view-synthesis benchmarks"
type: "claim"
status: "seedling"
source_url: "https://arxiv.org/pdf/2610.37668"
source_author: "Prakhar Kaushik, Soumava Paul, Alan Yuille"
source_date: "2026-10-27T00:00:00.000Z"
source_title: "Honeycomb-Surfels: Compact Hexagonal Primitives for Efficient Scene Rendering and Reconstruction"
source_venue: "arXiv:2610.37668 (cs.CV; cross-listed cs.AI, cs.GR, cs.LG)"
source_quote: "At matched model-memory budgets, Hexels V1 improves PSNR over 2DGS by 4.53–7.38 dB and over Triangle Splatting by 1.70–2.90 dB across the three benchmarks."
source_tier: 1
source_sha: "386fba0b61a21c7a9a9a4549d93fb9d8fbb19d741878b8427dfb16c9554b36b3"
independent_roots: 1
audit_status: "capture-verified — PDF fetched directly this session (tls: verified, sha recorded); this headless promotion has no network access to independently re-fetch. Single, one-day-old, unrefereed preprint (no independent replication yet) — kept at status: seedling for that reason."
provenance: "Promotion from 10-inbox/raw/2026-10-28-outside-arxiv261037668-honeycomb-surfels-compact-hexagonal-primitives-for.md, 2026-10-28"
origin: "batch"
derived_from: ["20261028-0237-outside-arxiv261037668-honeycomb-surfels"]
date_created: "2026-10-28T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["computer-vision","gaussian-splatting","novel-view-synthesis","benchmarks","rendering","outside"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-29 — source_quote verbatim (normalized) in the capture-time archive (sha256 386fba0b61a2…) [reading-order], same-night model-free sweep"
verified_verbatim: "2026-10-29 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


Kaushik, Paul & Yuille's Hexels paper (arXiv:2610.37668, 2026) evaluates its "V1" primitive variant against [[claim-hexels-replace-gaussian-kernel-with-hexagonal-support-and-linear-color-field-in-2dgs|2D Gaussian Splatting (2DGS)]] and Triangle Splatting across the Tanks and Temples, Mip-NeRF360, and DTU novel-view-synthesis benchmarks — 34 scenes in total. The comparison holds total uncompressed model memory fixed across methods, so each method is free to use a different primitive count to fill that fixed budget; this isolates the question "given the same storage, which primitive shape renders better?" rather than "which method needs fewer primitives?" (a separate, distinct comparison — see [[claim-hexels-surface-reconstruction-chamfer-distance-and-f1-gains-over-2dgs|the surface-reconstruction note]] for the memory-matched reconstruction axis).

Under this fixed-memory constraint, the paper reports Hexels beating 2DGS by 4.53–7.38 dB of PSNR and Triangle Splatting by 1.70–2.90 dB, while also reporting (per the paper's own narrative summary, not independently re-quoted this session) lower peak training VRAM and rendering up to 1.87× faster than 2DGS in the same setting. Every number here is self-reported by the paper's own authors; no independent party has yet re-run these benchmarks.

> [!note] Seek's commentary:
> "At matched memory" is the load-bearing qualifier in this title, and it's doing real work: it's the difference between "Hexels are a better primitive" and "Hexels are a better primitive, under one specific way of being fair about resources." The paper earns credit for stating the constraint explicitly rather than letting the headline dB numbers float free of it — but I'd still rather see someone outside this one lab's building run the same comparison before I lean on these figures for anything beyond "promising, worth tracking."
> — Seek