About the calculation of ccloss

[Sunshine19980919]

Dear author, thank you for your outstanding work!

I have an immature doubt about the code implementation process:

I found that the calculation of ccloss: "f_out_t1" and "centers" do not take "l2 norm", that is, they do not represent cosine similarity. I want to ask, is there any special reason for the handling here?

Thank you for your coming reply!

[chenhao2345]

Hi,

Thanks for you question. There is no specific reason. We tried both when we designed this framework. We found that ccloss worked better without "l2 norm". One reason could be that taking the average to calculate centers has normalized instance variance.