For full datasets, please see the individual dataset pages.

Sample Datasets

Sample datasets containing small amounts of data from all sensors (all starting from the same location) can be downloaded here:

Oxford RobotCar Dataset SDK

The core software needed to parse data in the Oxford Radar RobotCar Dataset, as well as example usage, has newly been included in the original Oxford RobotCar Dataset SDK. So in addition to previously available MATLAB and Python functions for loading and displaying data, including Bayer demosaicing and undistorting images, the SDK now includes MATLAB and Python functions for:

  • Parsing raw Radar data - LoadRadar and radar.load_radar
  • Parsing raw Velodyne data - LoadVelodyneRaw and velodyne.load_velodyne_raw
  • Parsing binary Velodyne data - LoadVelodyneBinary and velodyne.load_velodyne_binary
  • Converting polar radar data to Cartesian form - RadarPolarToCartesian and radar.radar_polar_to_cartesian
  • Converting raw Velodyne sensor data into a pointcloud - VelodyneRawToPointcloud and velodyne.velodyne_raw_to_pointcloud

As well as:

  • Example radar data viewiers - PlayRadar and play_radar.py
  • Example Velodyne data viewers - PlayVelodyne and play_velodyne.py
  • Updated extrinisics for new sensors (radar.txt, velodyne_left.txt, velodyne_right.txt)
  • Updated BuildPointcloud and build_pointcloud.py for handling newly included Velodyne HDL-32E data

The Oxford RobotCar Dataset SDK can be found here:
https://github.com/ori-mrg/robotcar-dataset-sdk

Oxford Radar RobotCar Dataset SDK

Building on top of the base SDK we provide additional specialisations for the new dataset which can be found here: https://github.com/dbarnes/radar-robotcar-dataset-sdk

Which includes:

Download Script

A python download script for the Oxford Radar RobotCar Dataset which:

  • Avoids the need for manually selecting individual files for download from the website
  • Allows filtering by dataset
  • Allows filtering by sensor

Optimised Deep Learning Data Loaders - coming soon…

To aid development we also plan to provide optimised example parsers, dataloaders and usage for using the radar data and ground truth odometry in Tensorflow, PyTorch and in OpenCV. If you would like access to these more urgently please get in touch and we will prioritise their release.