search

cvqluu / Angular-Penalty-Softmax-Losses-Pytorch

Angular penalty loss functions in Pytorch (ArcFace, SphereFace, Additive Margin, CosFace)
MIT License
481 stars 91 forks source link

The error of the training about arcface #15

Open LONG520520 opened 2 years ago

LONG520520 commented 2 years ago

hello： When I was training arcface's second epoch, what was the reason for this? image This is the visual diagram of training: image

msalman95 commented 2 years ago

I have the exact same issue, the Loss either goes to 0 or to NaN for arcface. Did you find the the solution for this?

EduanKK commented 1 year ago

I have the exact same issue, the Loss either goes to 0 or to NaN for arcface. Did you find the the solution for this?

I have same issue. In all arcface, sphereface, cosface loss.