MulRan Dataset for Urban Place Recognition

Highlights of the dataset


    1. Multimodal range sensors: Radar and LiDAR → for robust structural place recognition algorithms
    2. Multiple cities (multi-environments) → for environment-free loop detection algorithms
    3. Multiple times with month-level temporal gaps → for long-term localization research
    4. Multiple loop candidates, particularly
      • Reverse revisits → for rotation (viewpoint) invariant loop detection research
      • Revisits with lane-level differences → for translation change-robust loop detection research
    5. 6D baseline trajectories are provided for all sequences → not only place recognition, but also for odometry or SLAM studies.

The sensor suite

    • A single radar
    • A single LiDAR


MulRan's sequence environments and trajectories

Real scale


Detailed Highlights

  • Revisits with reverse direction and lane-level differences
    • Example: Riverside 01, 02, and 03

Data Format (TBA)

    • LiDAR
      • binary (same as KITTI)
    • Radar (polar image)
      • a single scan with a single time (at the last ray)
    • Radar (ray)
      • each ray has corresponding time