⇈ Download data
The copyright of the 3DCars dataset is owned by The Centre for Vision, Speech and Signal Processing, University of Surrey, UK. Permission is hereby granted to use the 3DCars dataset or its part for academic purposes only, provided that it is referenced in publications related to its use as follows:
K. Lebeda, S. Hadfield, R. Bowden. 2D Or Not 2D: Bridging the Gap Between Tracking and Structure from Motion. In Proceedings of the Asian Conference on Computer Vision, 2015.
⇈ Submit results
To be placed on the leader-board, please email us your results as returned by the evaluation script. Please copy&paste the results from your Matlab window into the email body and/or attach the created result file. In case you don't get a response within a few days, feel free to urge us.
If you want to have your submission marked as "confirmed", please send us the source code or binary of your implementation for a check. Your implementation will be used solely for the purpose of validating your benchmark entry; should you wish to discuss this issue, please reach us on the aforementioned address.
⇈ Leader Board
The following table shows quantitative results on the 3DCars dataset, given the ground truth. The tabulated values (for every tracker and sequence combination) are localisation error (distance of bounding box centres, in pixels) and overlap (of bounding boxes, in percent points).
|Conf.||Tracker||2D / 3D||Doughnut||Rally-Audi||Rally-Lancer||Rally-Opel||Rally-OTS||Rally-VW||TopGear 1||TopGear 2||Average|
1)TLD on Rally-Audi reported loss of tracking after only 3 frames.