牛津日夜数据集
收藏资源简介:
牛津日夜数据集是一个大规模的、以人为中心的3D数据集,旨在用于新型视图合成(NVS)和视觉重新定位,特别是在具有挑战性的照明条件下。该数据集弥补了现有数据集在真实头动、色彩和全日光照变化等方面的不足,支持两个核心基准:NVS和重新定位。数据集覆盖了超过30公里的记录轨迹和40000平方米的区域,为以人为中心的3D视觉研究提供了丰富的数据基础。该数据集利用Meta ARIA眼镜捕获以人为中心视频,并应用多会话SLAM技术来估计相机姿态,重建3D点云,并对在变化的光照条件下捕获的序列进行对齐,包括白天和夜晚。
The Oxford Day-Night Dataset is a large-scale, human-centric 3D dataset developed for novel view synthesis (NVS) and visual relocalization, especially under challenging lighting conditions. It addresses the limitations of existing datasets in terms of realistic head movements, color variations and full-day illumination changes, and supports two core benchmarks: NVS and visual relocalization. The dataset spans over 30 kilometers of recorded trajectories and covers an area of 40,000 square meters, providing a rich data foundation for human-centric 3D vision research. This dataset captures human-centric videos using Meta ARIA glasses, and leverages multi-session SLAM technologies to estimate camera poses, reconstruct 3D point clouds, and align sequences captured under varying lighting conditions, including daytime and nighttime.

- 1Seeing in the Dark: Benchmarking Egocentric 3D Vision with the Oxford Day-and-Night Dataset牛津大学 · 2025年



