Reproduce “Learning to Discriminate Scene Structures Makes Self-Supervised Depth Learning Scalable” 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
Mengtan Zhang, Yuchuan Cui, Yu Ma, Zizhan Guo, Wei Ye, Rui Fan
MaSS is a self-supervised monocular depth system that separates scene layout features from contextual features and trains layout representations to group similar predicted depth distributions. On four mixed-training evaluation datasets, it reports the best self-supervised results, including nuScenes Abs Rel 0.134 and δ₁ 0.852, and generalizes to eight unseen datasets. Training uses 150K images; larger-scale behavior remains open.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.