LoGeR – 3D reconstruction from extremely long videos (DeepMind, UC Berkeley)

Automated excerpt

LoGeR scales feedforward dense 3D reconstruction to extremely long videos. Even without backend optimization, LoGeR maintains strong geometric coherence and reduces scale drift over kilometer-scale trajectories. Both Pi3-Chunk and LoGeR strongly outperform prior feedforward baselines, and LoGeR* achieves the best average ATE at 18. 65.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.