Reproduce “SSLayout360: Semi-Supervised Indoor Layout Estimation from 360-Degree Panorama” 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
Phi Vu Tran
SSLayout360 introduces a semi-supervised learning framework for 3D indoor layout estimation from 360-degree panoramas, effectively reducing the need for extensive labeled data. It integrates the Mean Teacher framework with HorizonNet, achieving performance comparable to fully supervised methods using as little as 12% of the original labels on complex datasets like MatterportLayout.
Key insights
- Semi-supervised learning, specifically consistency regularization, can be effectively adapted for complex, fine-grained geometric regression tasks like 3D indoor layout estimation.
- Utilizing unlabeled data, even from auxiliary sources or different domains (like synthetic data), substantially reduces the reliance on manual annotations while maintaining or improving model performance.
- A carefully designed compound loss function with compatible supervised and unsupervised components (e.g., L1/L2 losses allowing a fixed consistency weight) simplifies hyper-parameter tuning and enhances training stability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.