MRP Drone Datasets

We present here MRP Drone Datasets 01 and 02 for academic use only.

All other use prohibited without express consent and licensing from the author.

Dataset 02 has been updated to include previously missing flights due to a corrupt upload file. The full file now includes 3 additional flights

Download MRP DATASET 01 (6.32gb) Download MRP DATASET 02 (4.80gb 5.58gb)

Questions or comments on this download?



MRP DATASET 01

  • "Investigating Open-World Person Re-identification Using a Drone", R. Layne, T.M. Hospedales and S. Gong. Workshop on Visual Surveillance and Re-identification, ECCV, Switzerland, 2014
  • Comprising 10 'flights':

    • Flight:DetctsUn.IDsUn.DetGallry
      FLYSQ1:7629330108675207
      FLYSQ2:661781076510
      QMFLY2:84381108428
      QMFLY3:66844586626
      curve2:25405922448
      indoor11:2336108125
      indoor21:4717191280
      libsq_:345991583301
      outdoor1:79710330467
      qmod1_:807841397939
      ------------------------------
      TOTAL :113610842279111331

  • Including ICT HSV and ELF features, cropped images at 48x128, detection confidences and raw Drone sensor data (uncalibrated altitude, yaw, pitch roll, height, battery, motor speed, etc)
  • 1Annotation labels for these datasets are synchronised, permitting inter-flight re-identification
  • We would like to host your annotations here, please provide a .mat file of the same format as annotations.mat, with your new annotations and email a zip or dropbox link to r.d.Click here if you're human@qmul.ac.uk and we will host a single merged file for each flight.
  • Example scripts included to help navigate the data structures

MRP DATASET 02

  • "Investigating Open-World Person Re-identification Using a Drone", R. Layne, T.M. Hospedales and S. Gong. Workshop on Visual Surveillance and Re-identification, ECCV, Switzerland, 2014
  • Raw frame data from 11 MRP Drone flights used in our work
  • Frame data corresponds to Drone sensor data in MRP Dataset 01, facilitating use cases such as extracting your own person detections, scene activity understanding, and many other possibilities.
  • 640 x 360 RGB images, ~16,000 frames total over outdoor and indoor environments with variable public crowding.