Reproduce “AirPlanes: Accurate Plane Estimation via 3D-Consistent Embeddings” 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
Jamie Watson, Filippo Aleotti, Mohamed Sayed, Zawar Qureshi, Oisin Mac Aodha, Gabriel Brostow, Michael Firman, Sara Vicente
Niantic researchers introduce AirPlanes, a method that combines robust 3D geometry with learned 3D-consistent embeddings to accurately estimate planar surfaces from sequences of posed RGB images. The system, capable of real-time operation, improved plane segmentation and fidelity metrics on the ScanNetV2 dataset while outperforming existing multi-view learning approaches.
Key insights
- The learned 3D-consistent embeddings effectively integrate geometric and semantic information, allowing the system to differentiate semantically distinct planar instances that purely geometric methods might incorrectly merge.
- A strong geometric baseline (3D reconstruction plus RANSAC) can achieve competitive performance, sometimes outperforming prior learning-based methods, highlighting the importance of high-quality initial geometry.
- The proposed 3D embeddings are modular and can enhance various existing 3D geometry estimators, consistently improving plane estimation metrics when integrated into different reconstruction pipelines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.