Improve your state of the art by using best meta optimizer and best activation function

[LifeIsStrange]

You could increase SMIM accuracy by using Ranger, which combine state of the art optimizers + gradient centralization
https://github.com/lessw2020/Ranger-Deep-Learning-Optimizer

Hortogonally, you would probably benefit from Mish too instead of the one you use (SGD ?) but should be tested after Ranger as it could regress accuracy (even if unlikely)
https://github.com/digantamisra98/Mish

You can expect from this big accuracy gains.
That would make your model even closer from perfection :)

[michalivne]

@LifeIsStrange thanks for your suggestions. We will look into them.

Cheers