Mask loss is 0.0 and overall loss doesnt converge && what is relevance of dt_num_classes ?

[Raj-08]

Hi ,
I am training segfix model and my mask loss is 0.0 for most of iterations. Also the loss doesnt converge.

Another question i have is what is relevance of dt_num_classes? is it related to the actual total number of classes?

[xiamibudayang]

hello, I am training segfix model too and I also meet the seem problem as the same as you. Have you find the answer now?
Thanks

[Raj-08]

Yes , the problem lies in your creation of direction maps . Visualise them and see if it makes sense as per the direction maps shown in research paper.