Reproduce “RoMe: Towards Large Scale Road Surface Reconstruction via Mesh Representation” 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
Ruohong Mei, Wei Sui, Jiaxin Zhang, Xue Qin, Gang Wang, Tao Peng, Cong Yang
A framework, RoMe, reconstructs large-scale 3D road surfaces from vehicle-mounted camera images and semantic annotations into meshes that integrate elevation, color, and semantic attributes. This method enables efficient auto-labeling and demonstrates robust performance for autonomous driving, processing 600x600 square meters in 2 GPU hours.
Key insights
- Optimizing road surface elevation using an MLP embedded within a mesh representation yields smoother and more accurate 3D structures compared to independent parameter optimization.
- Waypoint sampling provides a computationally efficient "divide-and-conquer" approach, making large-scale road reconstruction feasible with standard GPU resources by localizing optimization.
- Simultaneous optimization of mesh properties and camera extrinsic parameters is crucial for achieving high-quality, robust reconstructions, particularly in scenarios with imperfect initial calibration and dynamic scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.