Insta360 Open-Sources PanoLOG: Breaking Large-Scale 3D Reconstruction Bottlenecks with Panoramic Cameras
机器之心 · wechat · 2026-07-27
Insta360 Research Institute, collaborating with universities, proposed PanoLOG, the first large-scale outdoor 3D reconstruction framework for panoramic input, alongside the open-sourced Pano360 benchmark.
Traditional city-scale 3D reconstruction relies on numerous narrow FoV photos, making data collection expensive. Panoramic cameras reduce collection costs with a 360° view, but their "visible everywhere" nature causes traditional visibility-based partitioning strategies to fail. PanoLOG innovatively shifts the partitioning criterion from spatial visibility to gradient contribution, reallocating computation by evaluating camera optimization gradients to blocks, restoring the advantages of parallel training.
The framework also introduces an explicit sky sphere and panoramic depth supervision. It achieves state-of-the-art rendering quality on both proprietary and public datasets with a significantly smaller model size, providing a new path for low-cost city-scale 3D reconstruction and embodied AI.
More from Embodied
- AheadForm unveils an ultra-lifelike humanoid robot at WAIC 2026 in Shanghai — Olivier__OG · 2026-07-27
- Stanford AI Lab cites NVIDIA’s Alpamayo platform as a model for open Physical AI — StanfordAILab · 2026-07-27
- RoboDojo benchmarks 30 robot manipulation policies across 42 sim and 18 real tasks — jiqizhixin · 2026-07-27
- Neuralink's future: Brain implants could potentially eliminate back pain — DimaZeniuk · 2026-07-27
- τ(0)-VLA uses 40,115 hours of robot data to tackle long-horizon tasks — 机器之心 · 2026-07-27
- MotionBricks shows modular real-time motion generation for games, robots, and film — maier_ak · 2026-07-27