Clarification on ARCFACE loss function

Hi insightface team, thank you for this repo.

I had a question related to the arcface loss function.

In the 1st part of the denominator (j = y_i), you have added +m margin which increases the "Intra-class compactness"

But in the 2nd part of denominator (j != y_i), you are not adding or subtracting any margin.... I was just curious what would happen if we add -m margin to theta.... Won't that increases the "Inter class separability"

Thankyou for reading my question, I will be waiting for your response.