Reproduce “SAM 3D for 3D Object Reconstruction from Remote Sensing Images” 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
Junsheng Yao, Lichao Mou, Qingyu Li
This study systematically evaluates SAM 3D for 3D building reconstruction from monocular remote sensing images, demonstrating its superior performance over TRELLIS with an 8% reduction in FID and an 18% reduction in CMMD for individual buildings. The research also extends SAM 3D's capabilities to urban scene reconstruction through a segment-reconstruct-compose pipeline.
Key insights
- General-purpose image-to-3D foundation models, specifically SAM 3D, can effectively bridge the domain gap from natural images to monocular remote sensing imagery for 3D building reconstruction.
- SAM 3D consistently outperforms existing image-to-3D foundation models like TRELLIS in producing perceptually and semantically more accurate individual building reconstructions.
- While object-centric foundation models can be extended to urban scene reconstruction via composition, challenges such as computational cost, layout drift, and orientation ambiguities arise due to a lack of scene-level priors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.