Reproduce “Enforcing geometric constraints of virtual normal for depth prediction” 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
Wei Yin, Yifan Liu, Chunhua Shen, Youliang Yan
The authors train a monocular-depth network to estimate scene distances while preserving three-dimensional structure. The network matches pixelwise depth and the orientations of planes formed by distant groups of three reconstructed points. In a NYUD-Small ablation, virtual-normal loss lowers rel from 0.1427 to 0.1337 versus the weighted cross-entropy baseline; this result does not establish matched-data benchmark advantages.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.