Competition Rules

Dear participants, in order to create valuable results and provide a fair competition experience for all participants, please respect those rules:

  1. You can participate to 1, 2 or 3 tasks: they will be awarded separately.
  2. You must register and process the test set in time for your method to be evaluated.
  3. You must submit your method short description in time for your method to be included in the result paper.
  4. Your method must not include any manual annotation process.
  5. You must train on the training set or publicly available data exclusively.
  6. You must clearly indicate in your method presentation the training and validation sets you used.
  7. You should calibrate all your parameters using the validation set. Not doing so will be pointed out in the report.
  8. You may use the complete dataset for training. In particular, you can use the training sets for tasks 2 and 3 as unsupervised training examples for task 1. If you do so, please report it in your method's description.
  9. You must report in your method's description any redundancy in data that you make use of.
  10. We will not ask you to send code or binaries, but we encourage you to release your method under an open-source license after the competition.