RealEstate10K is a large-scale camera pose dataset containing approximately 80,000 video clips collected from approximately 10,000 YouTube videos, totaling 10 million frames. For each video clip, these camera poses form a track, where each pose specifies the camera's position and orientation along the track.

This dataset contains camera trajectories derived from video clips. These trajectories are obtained by running SLAM and bundle adjustment algorithms on a large number of videos.