(/!\ PR based off of the infer branch, not master. Should not impact this particular bit but still, be aware)
@51N84D (and others) I need you to be careful in reviewing this PR:
there's a new parameter called for_discriminator which I set to False in the gen's update and True in the disc's update
you have a global and a local loss. Not sure the vanilla way I updated to hinge loss makes the most sense so I need your input on this
especially, SPADE's hinge loss has an internal way of dealing with multi resolution so I think we can get rid of the for loop around the loss. If you agree I'll let you update
critical thinking's required here otherwise it'll be trained to nonsense ;)
(/!\ PR based off of the
inferbranch, not master. Should not impact this particular bit but still, be aware)@51N84D (and others) I need you to be careful in reviewing this PR:
for_discriminatorwhich I set to False in the gen's update and True in the disc's updatecritical thinking's required here otherwise it'll be trained to nonsense ;)