bhack
bhack
@fchollet @qlzh727 Can you migrate this to keras-cv now? It seems @LukeWood has not enough rights in this repo for the migration.
Is it binary for the library user? e.g. whole model compile or nothing?
But here we were talking about `BaseImageAugmentationLayer` not the whole model interface. In any case I don't know all the new refactors so do what you want with this ticket.
As `self.supports_jit` it is a bool how we control XLA vs `torch.compile` dynamo VS other compilers? A layer that could be compiled for one stack doesn't meant that it could...
Check also the thread at https://github.com/facebookresearch/segment-anything-2/issues/186
As you don't have a minimal example/dummy network to reproduce it can you check if it could be the same as https://github.com/tensorflow/tensorflow/issues/50400?
Please check https://github.com/tensorflow/tensorflow/issues/50400#issuecomment-1231502107.
> and indicate I cannot do anything Who is claiming this? I have proposed one popular solution in that comment that it is adopted in many implementations when you are...
@yifeipet Currently we are not offering a SSIM loss in TF directly but just an operator. I suggest to request or contribute a SSIM/MS-SSIM loss in https://github.com/keras-team/keras-cv if they are...
Any review on this?