Why the correlation is used to calculate the loss instead of the Frobenius Norm ?

[hsavatierdupre]

Hello,
I am trying to reproduce your code and I saw that you used the result of the correlation to calculate the distance used in the loss instead of the Frobenius norm as you indicate in the paper. Is it a mistake or voluntary ? And if voluntary why do you use it ?
Beside, I see that you use an argmax function to find the orientation after the correlation calculation but if I'm not mistaken, this function is not differentiable and therefore blocks the gradient descent. How did you deal with this ?
Thank you in advance!