Scal3R (ECCV 2026): online 3D reconstruction collapses on long videos due to global pose head
CMHungSteven · x · 2026-09-07
New ECCV 2026 paper Scal3R investigates why online 3D reconstruction models collapse on long videos even when per-frame depth remains stable. The authors find local geometry is often still reliable; it's the global pose head that breaks, and propose a fix accordingly.
More from Research
- PLANET lifts multi-object tracking into 3D scene geometry, hits SOTA on DanceTrack, SportsMOT & BFT — lealtaixe · 2026-09-07
- One author per paper: AI conference registration squeeze sparks authorship jokes — andimarafioti · 2026-09-07
- COLT 2027, the 40th Annual Conference on Learning Theory, Heads to Tokyo — FrnkNlsn · 2026-09-07
- HoloWorld unifies indoor-outdoor 3D urban generation with cross-scale context — Xiaobin Huang · 2026-09-07
- Dev Open-Sources Design of an LLM Memory Benchmark With Stale-Fact Scoring and Noise Scaling — True_Mongoose_7073 · 2026-09-07
- Engineer's 3-layer MLP classifies automotive radar targets, Macro F1 hits 0.764 — bruno_pinto90 · 2026-09-07