UniMetric: Training-Free Metric Scale Recovery for Geometry Foundation Models via VLM Size Priors
zhenjun_zhao · x · 2026-10-21
UniMetric is a universal, training-free framework that recovers metric scale for visual geometry foundation models (VGGT, π³, Depth Anything 3, Scal3R, etc.). It elicits uncertain object-height priors from a VLM, measures reconstructed heights along support-surface normals, and reconciles the resulting scale intervals via a convex objective into a single sequence-level scale — no extra sensors or metric supervision. On MetricBench-style driving/embodied-AI benchmarks, adding UniMetric cuts metric depth and reconstruction error by large margins (e.g., VGGT KITTI Chamfer distance 23.09m → 6.79m).
More from Research
- New paper makes 9 falsifiable predictions toward a major AI-enabled terror attack by 2027-2028 — StephenLCasper · 2026-10-23
- HumanoidPose recovers humanoid robots' full-body joint states from RGB video alone — cholz · 2026-10-23
- Sophont's Holon-S1 predicts genes and biomarkers from routine pathology slides — iScienceLuvr · 2026-10-23
- Glasgow study: raised eyebrows and slow blinks make speakers look confident — Limor_Raviv · 2026-10-23
- IFM open-sources RL360, the agentic RL infra behind K2 Horizon's training — rupspace · 2026-10-23
- Wayfarer discovers RL options from a single experience stream, SOTA on hard Atari games — MarlosCMachado · 2026-10-23