Princeton365: A Diverse Dataset with Accurate Camera Pose
Researchers have introduced Princeton365, a new dataset featuring 365 videos with precise camera pose information. This dataset aims to improve SLAM benchmarks by balancing data diversity with accuracy, utilizing a novel collection framework with calibration boards and a 360-degree camera. Princeton365 includes synchronized RGB and IMU data for indoor, outdoor, and object scanning scenarios. Additionally, a new scene scale-aware evaluation metric is proposed to better compare SLAM method performance across different scenes and identify failure modes. AI