tomealbuquerque / ordinal-losses Goto Github PK

Hi,

Congratulations for this nice work! I am trying to use CO2 and HO2 in my own work. I have two questions at this point:

1. In the paper, lambda term is multiplied with the ordinal loss terms (neighbor_loss), however, in the repository it is multiplied with cross-entropy or Entropy loss. It is a difference between the paper and the implementation right?
2. My dataset is large and it is not feasible to tune the lambda in a short time. To begin, can you write the best/default lambda as you did for the omega term?

Thank you.