GEM-Occ: Gaussian memory boosts indoor 3D mapping across rooms

A team of nine researchers from several institutions (including Hu Zhu, Bohan Li, Xianda Guo, and others) introduced HIOcc, a hierarchical indoor occupancy benchmark that unifies three major datasets—ScanNet, ScanNet++, and Matterport3D—under a common sparse semantic occupancy format while preserving original observation geometries like perspective RGB-D frames and pano-centric groups. HIOcc supports three evaluation regimes: local semantic occupancy prediction, room-level online occupancy mapping, and building-level mapping across connected panoramic environments. This fills a gap in existing benchmarks that focus only on single-view prediction or room-level perception, leaving long-horizon semantic mapping across connected indoor spaces underexplored.

Alongside HIOcc, the authors propose GEM-Occ (Gaussian Evidence Memory for semantic occupancy mapping). Instead of using pointmaps as persistent map states, GEM-Occ treats local visual geometry predictions as transient evidence, converts them into semantic Gaussian occupancy evidence and free-space ray evidence, and fuses them into a persistent hierarchical memory through visibility- and uncertainty-aware causal updates. The memory is organized into local caches, room-level submaps, and a building-level graph, and can be queried at any time via Gaussian-to-occupancy splatting. Experiments on HIOcc demonstrate improvements over prior indoor occupancy and Gaussian-based mapping baselines in local occupancy prediction, online map stability, free-space reasoning, revisit consistency, and building-level scalability.