Reproduce “When Wider Views Fail: Stress-Testing Feed-Forward 3D Reconstruction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Daisy Li, Kyle Gao, Quanyun Wu, Hanna Chomko, John S. Zelek, Jonathan Li
The authors test sparse-image 3D reconstruction under a fixed budget of ten images, varying the camera orbit from 30° to 360° across two urban scenes. They compare VGGT, Pi3X, and VGGT-Ω, evaluating surface agreement while retaining displaced predicted copies. At 360° in Scene II, VGGT-Ω reaches a 94.74% F-score, versus 15.26% and 9.44%, but against a model-generated pseudo-reference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.