Princeton365

Princeton365是一个包含365个视频的大型多样化数据集，这些视频具有准确的相机姿态。我们的数据集通过引入一种新颖的地面真实收集框架，弥合了当前SLAM基准测试在准确性和数据多样性之间的差距，该框架利用校准板和360°相机。我们收集了室内、室外和物体扫描视频，具有同步的单目和立体RGB视频输出以及IMU。我们进一步提出了一种新的场景规模感知SLAM评估指标，该指标基于相机姿态估计误差引起的运动流。与当前指标相比，我们的新指标允许比较不同场景中SLAM方法的性能，而不是像平均轨迹误差（ATE）这样的现有指标，这允许研究人员分析他们方法的失效模式。我们还提出了一个新的新颖视图合成基准，涵盖了当前NVS基准未涵盖的情况，例如具有360°相机轨迹的完全非朗伯场景。请访问princeton365.cs.princeton.edu以获取数据集、代码、视频和提交。

Princeton365 is a large-scale diverse dataset comprising 365 videos with accurate camera poses. Our dataset bridges the gap between accuracy and data diversity in current SLAM benchmarks by introducing a novel ground-truth collection framework that leverages calibration boards and 360° cameras. We collected indoor, outdoor, and object scan videos with synchronized monocular, stereo RGB video outputs and IMU data. We further propose a novel scene-scale-aware SLAM evaluation metric based on motion flow induced by camera pose estimation errors. Compared to existing metrics such as the Average Trajectory Error (ATE), our new metric enables performance comparison of SLAM methods across different scenes, allowing researchers to analyze the failure modes of their approaches. We also introduce a novel view synthesis (NVS) benchmark that covers scenarios not addressed by current NVS benchmarks, such as fully non-Lambertian scenes with 360° camera trajectories. Please visit princeton365.cs.princeton.edu to access the dataset, code, videos, and submission portal.