Any3R: Feedforward 3D Reconstruction from Any Camera Type
arXiv:2610.36150v1 Announce Type: new Abstract: Geometry foundation models recover camera poses and dense geometry from a set of images with a feed-forward network, but they expect pinhole images. That assumption is inside the backbone too: the positional encoding and the training loss both bind an…
Read the full story at arXiv cs.CV ↗
Timeline · 1 report
- 2026-10-28 04:00 · arXiv cs.CV
Any3R: Feedforward 3D Reconstruction from Any Camera Type