空间智能新作，影石开源户外全景重建算法

TL;DR - Insta360 and university collaborators introduced and began open-sourcing PanoLOG, a large-scale outdoor 3D Gaussian reconstruction framework for panoramic imagery. It lowers capture and model costs while improving rendering quality, supporting applications in embodied AI and spatial computing.

• PanoLOG replaces visibility-based scene partitioning—which breaks with 360° cameras—with camera-to-region assignment based on geometry and gradient contribution.
• Its two-stage pipeline combines global coarse training with panoramic monocular depth supervision, followed by parallel block-level refinement and merging.
• On Pano360, Ricoh360, and 360Roam benchmarks, it achieved leading rendering metrics while using smaller models than evaluated baselines.
• The team is releasing code, models, and Pano360, its new large-scale panoramic outdoor reconstruction benchmark.