AIM 2020 Efficient Super-Resolution Challenge Forum

Go back to competition Back to thread list Post in this thread

> About reproduction of winner methods

Dear Organizers:
Since the ground truth of validation set will be public availiable, we hope to verify that the winner methods can be reproduced WITHOUT training on the validation set. Therefore, we hope the winner team can release their training codes (including training scripts) to make sure that all the others can reproduce their results.

PS. According to the comments of the originazer to question "performance for testing phase", "For a fair comparison, you should not use validation dataset during training".


Posted by: Airia_CG @ July 9, 2020, 5:25 a.m.

The participants should also report whether they use the validation datasets during training.

Posted by: cszn @ July 9, 2020, 6:25 p.m.

Dear organizers:
Since training on the validation set can be the game-changer in this competition, we still hope all the other participants could reproduce the training process of the winner solution.
Thanks for your tremendous efforts!

Posted by: Airia_CG @ July 10, 2020, 12:41 a.m.
Post in this thread