Paper Proposes Adaptive Semantic Gaussian Allocation to Improve 3D Occupancy Prediction论文提出自适应语义高斯分配法，改进3D占用预测
S 1.7T11 sources1 个来源R7-research
Kanglin Ning and six co-authors (Yiran Zhao, Wenrui Li, Houde Quan, Qifan Li, Xingtao Wang, Xiaopeng Fan) posted "Learning Adaptive Semantic Gaussian Allocation for 3D Occupancy" to arXiv (arXiv:2607.21896v1, cs.CV) on July 24, 2026.
The paper targets semantic 3D Gaussian representations, which render semantic primitives into a voxel volume under voxel-wise supervision for 3D semantic occupancy prediction; it notes that while recent methods improved primitive shape flexibility, geometry-guided initialization, and progressive densification, none has explicitly addressed how to select the most useful Gaussians, the gap this paper's adaptive allocation approach targets.
D semantic occupancy prediction underpins autonomous-driving perception stacks, so an allocation method that concentrates limited Gaussian primitives on the most informative regions could improve real-time occupancy accuracy under the compute budgets of onboard AD/ADAS systems.
宁康林（Kanglin Ning）等7位作者于2026年7月24日在arXiv发表论文《Learning Adaptive Semantic Gaussian Allocation for 3D Occupancy》（arXiv:2607.21896v1，cs.CV分类），合著者包括Yiran Zhao、Wenrui Li、Houde Quan、Qifan Li、Xingtao Wang、Xiaopeng Fan。