UBC and Sony release Solv3R, turning a looped Transformer into a reusable 3D geometry solver
CSProfKGD · x · 2026-11-06
Researchers from UBC, Wayve Labs, SFU, Toronto and Sony unveiled Solv3R, asking whether geometric foundation models can move beyond fixed-set prediction and solve iteratively like classical SfM. The key idea: turn a looped Transformer into a reusable learned geometric solver for multi-view 3D reconstruction.
- Paper on arXiv since Nov 2, 2026
- Inference/eval code and pretrained weights released Nov 5, 2025 under Apache-2.0
- Offline and streaming benchmark evaluation code to follow
More from Research
- MIT's 12 Free AI Textbooks: Deep Learning, RL, Fairness All With Links — luisdans · 2026-11-06
- Quiz: can you identify these four classic attention patterns? — cneuralnetwork · 2026-11-06
- Laude awards four $10M Moonshots labs; Cornell team wins for civic discourse AI — 996roma · 2026-11-06
- Shadaj Lohia presents Hydro: chasing zero-cost correctness beyond C++, Zig and Go — ShadajL · 2026-11-06
- Lighting a Christmas tree with Stein Variational Gradient Descent — PTenigma · 2026-11-06
- Alex Wang's Perplexity reading group slides reframe Transformer design through associative memory — andrewgwils · 2026-11-06