Reproduce “GNFormer: Gaussian Neural Field Transformer for Efficient Camera-based 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

Guangcheng Chen, Li He, Hong Zhang

GNFormer predicts voxel occupancy and semantic labels from camera images. Gaussian primitives focus local feature aggregation near occupied regions, while an MLP decodes semantics from Gaussian features and voxel positions; regularization encourages support near occupied voxels. On Occ-ScanNet validation, it reports IoU 74.3, mIoU 66.9, and 30.8 FPS using 256 Gaussians, compared with reported methods.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.