Reproduce “SHINE-Mapping: Large-Scale 3D Mapping Using Sparse Hierarchical Implicit Neural Representations” 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
Xingguang Zhong, Yue Pan, Jens Behley, Cyrill Stachniss
SHINE-Mapping creates accurate and memory-efficient 3D maps of expansive outdoor environments by employing sparse hierarchical implicit neural representations and 3D LiDAR data. The system supports incremental map updates for continuous operation while effectively mitigating catastrophic forgetting.
Key insights
- Sparse hierarchical feature grids, combined with hash tables, provide a memory-efficient and scalable implicit representation for large-scale environments, overcoming the limitations of dense voxel grids.
- The use of Binary Cross Entropy loss on projected signed distances derived from LiDAR measurements offers a robust and sensitive supervision signal for learning SDFs, effectively emphasizing surface reconstruction.
- A regularization term based on importance weights effectively mitigates catastrophic forgetting in incremental neural mapping, allowing for continuous and consistent map updates over long operational periods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.