Skip to content

AURA — the trust layer for splats

A plain 3DGS/DBS checkpoint renders fast but ships no notion of per-primitive trust. AURA keeps those fast Gaussian / DBS-Beta renderers where they are strong and adds the layer they do not provide: a calibrated confidence every carrier carries, turned into a distribution-free certificate, turned into a certified streaming/LOD ladder — with that confidence travelling in every standard container (glTF, USD, SPZ), all of it gate-checked and CPU-reproducible.

The chain is Photogrammetry → NeRF → 3DGS → AURA: not a faster renderer, but a more trustworthy, inspectable asset on top of one.

Capability reel: calibrated confidence, certified LOD, interchange

Install

pip install aura-splat        # 1.1.0 · Python 3.11+
# or from source: pip install -e ".[dev,gpu,assets]"

Then follow Quickstart: fetch a scene, train carriers, render, export, ray-query.

Evidence (v1.1.0)

Area Result
Full suite 1928 passed, 37 skipped, 0 failed — CPU-reproducible
Certificate bounds 16/16 split-conformal bounds hold
LOD ladder Certified streaming ladder with calibrated per-carrier confidence
Interchange Confidence survives glTF, USD (primvars:aura:confidence), SPZ round-trips
Open artifacts DOI 10.5281/zenodo.21500723 · REPRODUCE.md replays the paper with no data or GPU

Scope (honest bounds)

No official-leaderboard SOTA claim anywhere in this repo. Open items are stated as open: a full 8-scene true-3DGS control, external reproduction, and a handful of demo-stage carriers. Negatives are kept, not hidden.

Guides

Start with Quickstart, then read the passes in order: calibrated confidence → distortion budget → cross-scene → full-res render loss → certified LOD → BVH ray query → carrier registry → relight decision.