Evaluation

The problem is a classification (action recognition) problem. A more detailed description could be found here.

There are 2 phases:

• Phase 1: dry-run phase. We provide you with labeled training data and a small batch of unlabeled dry-run (validation) data. You will receive feedback on your performance on the small validation set only during this phase. For submission, please STRICTLY name your submission .csv file as "track1_pred.csv" and submit a zip file that contains this submission .csv file. The performance of your LAST submission will be displayed on the leaderboard.
• Phase 2: testing phase. You can make predictions for the test dataset. Your last submission of phase 1 will be automatically forwarded. For submission, please STRICTLY name your submission .csv file as "track1_pred.csv" and submit a zip file that contains this submission .csv file. Your performance on the test set will appear on the leaderboard when the organizers finish checking the submissions.

This competition only allows you to submit the prediction results (no code), please STRICTLY name your submission .csv file as "track1_pred.csv" and submit a zip file that contains this submission .csv file:

However, the winner and the two runner-ups are required to submit their model codes. Challenge organizers will test the reproducibility. Failure of reproducing the performance will lead to the result on the leaderboard marked as invalid.

The submissions are evaluated using the Accuracy metric. For each video you would and would only provide ONE prediction label. The prediction is correct only if it is a full match with the ground-truth The metric is re-scaled linearly between 0 and 100, 0 corresponding to a random guess and 100 to perfect predictions.

UG2+ 2021: Rules

Submissions must be made before the end of phase 1. You may submit up to 10 submissions every day and 200 in total.

More detailed rules could be found here.