Reproduce “VGOcc: Learning Visual-Geometric Gaussians for Vision-Centric 3D Driving Occupancy Prediction” 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
Junhong Lin, Xianda Guo, Kangli Wang, Yuqi Ye, Xiaoyu Liang, Yanlun Peng, Wei Gao
VGOcc introduces Visual-Geometric Gaussians, a framework that leverages visual and geometric cues from foundation models to enhance vision-centric 3D semantic occupancy prediction. It achieved an SC IoU of 34.07 and SSC mIoU of 21.75 on the nuScenes dataset, outperforming prior sparse Gaussian and dense structured representation methods.
Key insights
- Leveraging foundation models for explicit visual and geometric cues in Gaussian initialization and refinement yields a more robust 3D scene representation.
- Spatially balanced and geometry-guided Gaussian initialization is crucial for effective scene-specific semantic priors and comprehensive coverage.
- Patch-local pose conditioning and cross-view feature aggregation are essential for resolving depth ambiguity and enhancing consistency during Gaussian refinement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.