Solv3R: looped transformer iteratively refines multi-view geometry across reconstruction tasks
Jeong-gi Kwak · hf · 2026-11-05
Researchers from UBC, Wayve Labs, Simon Fraser, Toronto and Sony propose Solv3R (arXiv, Nov 2 2026), a looped transformer that iteratively refines multi-view geometry from its current state.
- Core idea: iterative refinement instead of one-shot feedforward prediction
- Supports incremental updates and caching: handles incoming frames, evolving scenes, and unordered new views
- One set of weights covers multiple reconstruction tasks without specialized per-task modules
- Team spans UBC, Wayve Labs, SFU, University of Toronto, Sony Semiconductor Solutions and Sony Corporation
More from Research
- Template-based multilingual safety pipeline shows English-only evals misjudge alignment — swiss-ai · 2026-11-05
- 'Grafting' paper: SFT on donor checkpoints + weight merging beats on-policy self-distillation — AdtRaghunathan · 2026-11-05
- Physical Intelligence's RLT uses RL tokens to make robot fine manipulation 3x faster — svlevine · 2026-11-05
- FrankenSim's physics-based flute synthesis gets graded against real recordings — doodlestein · 2026-11-05
- Randomized Algorithms Everywhere: Why Randomness Beats Determinism in Production — arpit_bhayani · 2026-11-05
- CloudOpsBench: open-source benchmark tests AI agents on 45 real cloud ops scenarios — DarkMG1 · 2026-11-05