Losses are abnormal

[poodarchu]

after training the model as stated in readme, the classification loss is only about 0.009, and regression loss is about 1.0 at the beginning.

[qian-qi]

Please check your environment. The classification loss should be about 2 after initialization. It is directly from 4xlog(1+exp(6x0.5))/6=2.03.
After training, the classification loss should be similar to the regression loss.

[poodarchu]

[poodarchu]

I think it doesn't have relationship with the environment. and the modification is simple. So I think maybe you miss some details?

[qian-qi]

I find a potential problem in your code. The normalization term is from focal loss and we don't have it for the dr loss.

[poodarchu]

ok thanks.