Description of the algorithm

In addition to submitting annotation results, participants are required to upload a description of their algorithm. Ideally this document is a scientific publication or a technical report, with enough details to allow others to reimplement it. References to accessible literature are acceptable: there is no need to repeat information that can be found elsewhere. If you have not written a detailed paper yet or if you want to withhold detailed information about your method, please indicate the reasons for this in the description file you submit and describe the system only briefly, using the checklist given below:

  • Contact details.
  • Name of the algorithm.
  • Give the overall structure of the algorithm and briefly describe each step in this structure.
  • Does your system use additional training data? If so, describe the characteristics of that training data.
  • Do predictions for a given frame solely rely on that frame? Or do they also rely on previous frames? on following frames?
  • Give computation times (number of frames processed per second).
  • If the algorithm has been tested on other datasets, you could consider including those results.

Results without a suitable description will not be processed.

How Often can Results be Submitted?

To limit the chance of training on the test set, results from the same algorithm will be evaluated once a week at most. Only the last and the best submissions will be reported on the 'results' pageIf you want to evaluate two substantially different algorithms, you are allowed to submit a second set of results, along with a description, under a different algorithm name.


You can upload your automatic tool annotations and the description of your algorithm at the 'files' page. Please follow the formatting instructions given on the 'evaluation' page. In particular, make sure the zip archive is named after your algorithm's name.