Winner selection, 8x SSIM

Leaderboard rank looks sorted by only ‘8x’ SSIM.
Is it same at the final challenge?

The public leaderboard sorts by 8x and then 4x if 8x SSIM is equivalent.

The challenge will have a separate winners for 8x and 4x so this won’t be an issue. We’re releasing a post with more challenge details this week.

Thanks,
Tullie

1 Like

Hello. I was under the impression that winners would only be selected for each track, not for each acceleration of each track.
Does this mean that there are 4 winners or simply that the winner for each acceleration will also be mentioned in an announcement?

@tullie Also, I would be most grateful if the link to the mentioned announcement could be given.

Hi @veritas9872,

There will be 3 winners:

  • 1 for single-coil 4x
  • 1 for multi-coil 4x
  • 1 for multi-coil 8x

This is currently reflected on the challenge guidelines page.

The mentioned announcement is here: https://ai.facebook.com/blog/fastmri-challenge/. There will be another announcement before the end of the day when the dataset has finished uploading.

Thanks,
Tullie

@tullie I have read through the official rules and they stated that the third winner would be selected as follows.

“The most academically interesting Reconstruction from both Tracks combined presenting the most novel approach will also be selected as a potential winner.”

Am I correct in believing that the three winners will no longer be judged (mainly) by “academic interest”?

Also, I am curious as to whether there is no winner for single-coil 8x.

It’s possible that they’ll be an “academic interest” winner in the case of a draw. There is several reason’s we didn’t include single-coil 8x, one being that we have have limited invitations to speak at the NeurIPS workshop.

Thanks,
Tullie