Get Started
Research questionHow can online 3D reconstruction maintain globally consistent poses and geometry over long video sequences?Online reconstruction can remain locally accurate while its estimated global poses drift over long videos. Anchoring every pose to the first frame can push the model beyond its training distribution and cause geometric collapse.
Computer Vision
Image & Video Processing
Machine Learning
Latest papersRecent research connected to this question, newest first.Scal3R: Learning Efficient Multi-Relative Pose Query for Scalable Online 3D ReconstructionThis paper uses multi-reference relative pose estimation with a frozen reconstruction backbone and online pose-graph optimization with loop closure. It reports evaluations on several driving, synthetic, indoor, and dynamic-scene datasets; conclusions are limited to the reported sequential settings and do not establish performance for offline reconstruction or other scene and camera conditions.research paper · Sep 3, 2026
Related questions
How can camera-controlled video generation preserve spatial consistency over long horizons despite noisy 3D memory?How can sparse structure-from-motion remain reliable in repeated or symmetric scenes while limiting redundant computation?How can multi-view 3D reconstruction remain scalable without costly dense correspondences in online VSLAM and unordered images?How can dynamic 3D reconstruction handle severe temporal asynchrony across cameras during complex motion?
Home
Topics
Search
Library