Why the result with this implementation is better than reported one in the original paper?

[sooonwoo]

Hi, thanks for your impressive work!

It seems that we can get quite a lot better results with your implementation. Could you let me know what is the reason for the difference?

(Lb=40) 6.4% (Yours) | 13.81±3.37% (FixMatch(reported))
(Lb=20) 9% (Yours) | N/A. (FixMatch(reported))

[kekmodel]

I think it's weird, too. I think there is something wrong with my code.

[lingjieyi]

I think it's weird, too. I think there is something wrong with my code.

Did you only run seed=5? I got 85.24 by setting seed=1 and total_epoch = 512 running this code.