Reproduce “Sky2Ground: A Benchmark for Site Modeling under Varying Altitude” 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
Zengyan Wang, Sirshapan Mitra, Rajat Modi, Grace Lim, Yogesh Rawat
SkyNet combines ground, aerial, and satellite views for camera localization in multi-altitude 3D mapping. It uses separate satellite processing, masked attention, and curriculum training that shifts toward harder views. On 10 held-out sites, it reaches domain-averaged RRA@5 of 83.2 versus 73.6 for zero-shot VGGT, and RTA@5 of 66.5 versus 44.5. These are localization results, not evidence of improved rendered scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.