Anyone using Weighted Cross Entropy?

[sagarhukkire]

Hi All

As a part of assignment I am trying to use weighted cross entropy with Vnet, but I am stuck :( Here are my findings

1. Same hyper parameter but loss converged to 0.32
2. Then I increased learning rate still loss converged to 0.32
3. I changed learning policy from " step" to "inv" ,here my dice score increased but loss converged to 0.32 Any idea what's the issue ?

Thanks

[sagarhukkire]

Mentioned solution in Readme