← The frontier
Technology & AISep 3, 2026

Scal3R: Learning Efficient Multi-Relative Pose Query for Scalable Online 3D Reconstruction

Online 3D reconstruction models perform poorly on long videos.

Online 3D reconstruction models perform poorly on long videos. This happens because regressing poses relative to a fixed first-frame anchor forces extrapolation far beyond the training distribution. Small drifts accumulate and amplify into significant geometric collapse.…

The frontier is open to all. Sign in to learn this from first principles and save it to your knowledge base.