The account lyx1430243059 has submitted the test results to the Public Leaderboard，but got refused by the system (red cross),.

I want to know why I was rejected? Because I use Google Cloud Disk?, thank you.

Hello @3li, is this the EDDGAN submission? Your first few submissions didn’t contain a ‘reconstruction’ key, and I think the most recent one had a size mismatch (`(36, 320, 320, 1)`

vs. `(36, 320, 320)`

). Example usage of the `save_reconstructions`

function is here:

Note that if you’re submitting to the single channel knee leaderboard, you’ll need to add `_v2`

to each of your filenames before uploading your zip file.

I succeeded in showing the result on the public leaderboard, but it was obviously wrong. I think this is caused by normalization. But how can I deal with this problem? In my program, I normalize all pictures to a value between 0-1 and then enter the network for training. Is this wrong? I know that the predicted image between 0-1 should be returned to the original pixel range again. But I don’t know how to do it. I know my method may be a bit stupid. But what should be the correct normalization method? Please help me. Thank you. good luck.

No worries. I don’t think we apply any normalization. Every ground truth is simply the inverse Fourier transform of the k-space data, followed by square-root sum-of-squares coil combination. The SSIM function calculates the data range based on the ground truth, so values don’t need to be between 0 and 1. In terms of preparing your entry, this code:

is the exact code that I used to prepare an entry that achieved 0.7562 SSIM. I would just follow that for identifying discrepancies.

Is the log file necessary? I did not use pytorch for training, but referenced the ‘save_reconstructions’ function in the official github. Therefore, can I just store reconstructed images (number of slices, 320, 320) in h5 files, and then upload all the ‘_v2’ files?

Yes that is correct.