Google Research

Bike Video Dataset


This dataset contains 91,866 frames extracted from 11 videos. The videos were recorded using a hand-held phone camera while riding a bicycle. The dataset was primarily created to test the ability of self-supervised depth-learning methods to learn from videos with complex and non-smooth ego-motion.

  • 91,866 frames from 11 videos
  • Download size: 33.6 GB
  • No labels
  • Format: jpeg
  • Resolution: 1280x720
  • Frame rate: 30 fps
  • Device: iPhone 6
  • Location: Mountain View, CA, USA
  • Date: March 2018
  • Faces and licence plates have been blurred for privacy reasons.