3DGART: Particle-Centric Backprop Speeds Up Gaussian Ray Tracing Training 3-3.5x
zhenjun_zhao · x · 2026-08-19
arXiv paper 3D Gaussian Accelerated Ray Tracing: Fast training through particle-based backward propagation by Laurent Vit, Oliver Batchelor, and Richard Green.
- Bottleneck: Rasterization-based Gaussian splatting can't handle reflections/refractions/shadows well, but ray-traced Gaussians are costly to train — the bottleneck is the pixel-centric backward pass, where many threads atomically accumulate gradients into the same primitives, causing severe contention and serialization.
- Method: 3DGART reorganizes backward propagation around primitives: conservative perspective-correct screen-space bounds build a compact intermediate buffer and tile-primitive mapping, letting each thread accumulate one primitive's contribution across covered pixels — turning gradient computation from a contention-heavy scatter into a structured gather.
- Results: 3-3.5x raw training speedup over per-pixel baseline and 4x over 3DGRT on Mip-NeRF 360.
More from Research
- mini Jev: an open 1.05MB decision head on frozen Qwen3 0.6B outputs agent action probabilities — AlternativeSure2891 · 2026-09-21
- Weighting others by relatedness isn't new: paper models agents learning relational value — xuanalogue · 2026-09-21
- Paper critique: MLP size unaccounted for in Transformer logarithmic depth bounds — kfountou · 2026-09-21
- Robot Leaderboard Under Fire: 5 Trials Per Task, Authors Commit to 15 — YuXiang_IRVL · 2026-09-21
- Tabular foundation models enter the arena, mocked: 'wait until you find out about xgboost' — burny_tech · 2026-09-21
- LLMs can draw DAGs but can't guarantee d-separation, notebook discussion — fdellaert · 2026-09-21