-
Notifications
You must be signed in to change notification settings - Fork 8
Open
Description
Because the loss is \lambda D + R, when I even use \lambda=256 with MSE loss, the \lambda D part is much more than R part, which make bpp_mv_y be 0.
I wonder how can I solve th problem?
Besides, I follow the progressive training steps. But in step 1, I meet the problem.
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels