Reproduce “SG-BEV: Satellite-Guided BEV Fusion for Cross-View Semantic Segmentation” 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
Junyan Ye, Qiyan Luo, Jinhua Yu, Huaping Zhong, Zhimeng Zheng, Conghui He, Weijia Li
SG-BEV develops a framework for comprehensive urban building segmentation by combining satellite imagery and street-view perspectives. It achieves improved accuracy in detailed building attribute segmentation and robustness against occlusions and distance compared to existing single-view and cross-view methods.
Key insights
- Leveraging satellite-derived building footprints can effectively guide the reprojection of street-view features into a Bird's Eye View (BEV), resolving the uneven feature distribution common in traditional BEV methods.
- A combination of depth-aware processing and a sophisticated fusion mechanism can effectively integrate complementary information from diverse viewpoints (overhead satellite and ground-level street view).
- Addressing cross-view challenges like feature alignment and preserving architectural details requires dynamic and iterative fusion, alongside targeted reprojection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.