Graph3R: Unifying View Graph Initialization with Feed-Forward Reconstruction

Key points

• Feed-forward reconstruction recovers camera poses and dense geometry from image sets whose views are assumed to come from a single scene.1
• Reusing the multi-view features allows Graph3R to reconstruct disjoint scenes separately with a single backbone pass.1
• Regarding the multi-session mapping task, Graph3R improves cross-session localization and achieves competitive pose estimation accuracy, while delivering a runtime speedup of approximately $35.9\times$ over the baseline when reconstructing the query group.1
• By integrating learned view association with geometric inference, our approach takes a step toward end-to-end reconstruction from uncurated image collections.1

Sources (1)

• [1]Graph3R: Unifying View Graph Initialization with Feed-Forward Reconstruction
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 06:50 AM

  We introduce Graph3R, which unifies view graph initialization with feed-forward reconstruction through a shared geometric representation.

  Feed-forward reconstruction recovers camera poses and dense geometry from image sets whose views are assumed to come from a single scene.

Extractive summary: sentences quoted from the sources.