Confidence-aware densification beats 3DGS-MCMC across all budgets, recovering 0.84dB of 1.05dB PSNR gap
zhenjun_zhao · x · 2026-10-13
A team including 3DGS core authors (Bernhard Kerbl, Lukas Radl) proposes confidence-aware densification to fix a known flaw in 3DGS-MCMC: opacity-based sampling concentrates new primitives in well-reconstructed foreground, leaving backgrounds under-reconstructed.
- Keeps the fixed primitive budget and gradual growth, but places primitives by image error: new primitives spawn where screen-space gradients are large, and the least-contributing primitives are relocated to regions with low learned confidence
- Across 24 scenes, it beats 3DGS-MCMC on PSNR/SSIM/LPIPS at every budget from 100k to 2M primitives
- At 100k, with no prior or initialization stage, it recovers 0.84dB of the 1.05dB PSNR gap by which prior-free ConeGS leads 3DGS-MCMC
Figures show opacity-proportional sampling misses high-error regions that the confidence-weighted gradient distribution covers.
More from Research
- Allen AI's TutorMoments Benchmark Finds LLM Tutors Over-Help Students — allen_ai · 2026-10-13
- Ben Goertzel revisits Omohundro's Basic AI Drives, finding six assumptions that decide which minds apply — bengoertzel · 2026-10-13
- Sim-to-real skepticism: pixel-space native training beats generated simulators, argues DeWitte — chris_j_paxton · 2026-10-13
- DeepMind publishes survey mapping theories of machine consciousness — Dr_Atoosa · 2026-10-13
- Noise: a speedrun for pretraining transformers without backprop — madhavsinghal_ · 2026-10-13
- Harvard releases freeinference_agentic_trace, a million-scale agent trace dataset — harvardMadsys · 2026-10-13