2020 VIPriors Object Detection Challenge Forum

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

> About pre-trained model

Dear organizers,
Thanks for organizing VIPriors competitions.

I know that this challenge does not allow using any pre-trained checkpoint, including any pre-trained backbone. Then, to warrant the competitive integrity of the competition competitive participants may expect a request to share their code with the organizers for a reproducability study.

I wonder how code can tell that we didn't use such pre-trained models. I would like you to tell me how to find out whether or not pre-trained model is used by code.

Thanks,
Byeongjo Kim

Posted by: ByeongjoKim @ June 29, 2020, 2:02 a.m.

Hi Byeongjo Kim,

This is a good question, and an important one. First of all, this competition is an academic competition. We apply to the academic integrity of all competitors to adhere to the rules of the competition. There are also no major prizes to be won, so there should not be any significant incentive to cheat.

In the worst case we might have to use our reserved right to disqualify any submissions that we have reasonable doubt about adhering to the rules. We require code submission so that we could, only if we deem it necessary, attempt a reproduction of the submitted model. In that case of course we should be able to reproduce the submitted results using just the provided code, and without pre-trained models or extra data.

But again, we hope we can trust the academic integrity of all competitors.

Best,
Robert-Jan

Posted by: rjbruin @ June 30, 2020, 8:52 a.m.
Post in this thread