NeurIPS VisDA-21 Challenge Announcement - Updated Ensembling and Energy efficiency Rules

[usmn]

Dear participants,

Thank you for your continued interest in the VisDA-21 competition!

We would like to let you know that we took your feedback into account and updated rules to allow some ensembling, and will also take time required for training the model into consideration.

Ensembling: Ensembling is now allowed, but each additional forward pass through a parameter will count towards the total number of parameters allowed (100M). For example, an ensemble that passes through a 50M model twice will count as 100M towards the parameter cap.

Energy efficiency: Teams must report the total training time of the submitted model which should be reasonable (which we define as not exceeding 100 GPU days of V100 (16GB version) but contact us if unsure). Energy efficient solutions will be highlighted even if they do not finish in the top three.

If you have any questions, please contact us via visda-or...@googlegroups.com .

Good luck and have fun!

VisDA-21 team

http://ai.bu.edu/visda-2021/