Reproduce “HoliCity: A City-Scale Data Platform for Learning Holistic 3D Structures” 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
Yichao Zhou, Jingwei Huang, Xili Dai, Shichen Liu, Linjie Luo, Zhili Chen, Yi Ma
Researchers from the University of California, Berkeley, and Bytedance Research introduced HoliCity, a city-scale 3D data platform that combines high-resolution Google Street View panoramas with precise CAD models of downtown London to provide ground truth for learning holistic 3D structures. This resource enables superior performance and generalizability for urban 3D understanding tasks compared to models trained on existing synthetic or indoor datasets.
Key insights
- Leveraging precise CAD models and real-world imagery, combined with a robust alignment pipeline, offers a cost-effective and scalable method to generate high-quality, structured 3D ground truth for complex urban environments.
- The domain gap between existing synthetic or indoor 3D datasets and real-world outdoor scenes is substantial, making dedicated outdoor structural datasets like HoliCity crucial for developing generalizable 3D vision models.
- Directly learning human-interpretable, high-level geometric primitives from images can lead to more compact, intuitive, and efficient 3D scene understanding, moving beyond noisy low-level point cloud representations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.