Reproduce “Depth Anything in $360^\circ$: Towards Scale Invariance in the Wild” 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

Hualie Jiang, Ziyang Song, Zhiqiang Lou, Rui Xu, Minglang Tan

DA360 adapts a perspective-depth model to estimate depth from 360° panoramas. It learns a per-image disparity shift, enabling point clouds from disparity with one scale adjustment, and uses circular decoder padding to account for neighbors at panorama boundaries. On Detroit’s curated Metropolis benchmark, DA360 ViT-S reports AbsRel 0.1922 versus 0.2898 for retrained PanDA, under different alignment protocols; absolute scale remains unknown.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.