Hi all,
Is there any timeline for adding the EfficientNet pretrained models to the Hub in the SavedModel format that is compatible with TF2? Currently EfficientNet models are served only in hub.Module format which is not compatible with TF2.
Best regards
I'm wondering if this is something that the community can get done with a little guidance? A ton of these issues seem to rely on the mysterious "tensorflower" and go back couple years... so I'm just wondering if maybe with some pointers we can get this done. I tried to migrate the modules to TF2 but it seems there's embedded model metadata that explicitly prevents TF2 from accepting this.
Unfortunately, it is typically not possible to convert TF1 models to TF2 directly (at the least, fine-tuning would not work) and we have to wait until a respective team in TensorFlow or beyond decides to publish them. I can say that there are ongoing efforts in getting TF2 EfficientNet models, but no ETA.
Thanks for the response.
Hi @michal-kierzynka and @TechnikEmpire, these are now available here: https://tfhub.dev/s?tf-version=tf2&q=efficientnet
Will close this bug -- feel free to reopen if this isn't what you were looking for!
@j-grimstad Fantastic, thank you!
@j-grimstad thank you very much :)
Most helpful comment
Hi @michal-kierzynka and @TechnikEmpire, these are now available here: https://tfhub.dev/s?tf-version=tf2&q=efficientnet
Will close this bug -- feel free to reopen if this isn't what you were looking for!