Closed marcusturewicz closed 1 year ago
Any ETAs here?
Hi there,
Yes, we'd welcome such a contribution. However, we have no availability to work on it, so you'd have to open a PR yourself (we will review the PR and guide you). Do you have an ImageNet checkpoint available for the model?
Got it, i'll see if i can make a PR on this. i have some components built in the past but just need to clean it up a bit
@ypeleg any scope to contribute here?
Hello, Thank you for reporting an issue.
We're currently in the process of migrating the new Keras 3 code base from keras-team/keras-core to keras-team/keras. Consequently, This issue may not be relevant to the Keras 3 code base . After the migration is successfully completed, feel free to reopen this Issue at keras-team/keras if you believe it remains relevant to the Keras 3 code base. If instead this Issue is a bug or security issue in legacy tf.keras, you can instead report a new issue at keras-team/tf-keras, which hosts the TensorFlow-only, legacy version of Keras.
To know more about Keras 3, please read https://keras.io/keras_core/announcement/
Transferring from https://github.com/tensorflow/tensorflow/issues/47233 as requested
System information
Describe the feature and the current behavior/state.
DeepMind recently introduced their new "Normalizer-Free" NFNets, matching the test accuracy of EfficientNet-B7 on ImageNet while being up to 8.7x faster to train, and their largest models attain a new state-of-the-art top-1 accuracy of 86.5%. It seems fitting that these would be available in keras since they surpass EfficientNets which are already available.
Will this change the current api? How?
This would introduce new APIs i.e.,
tensorflow.keras.applications.NFNetF0
through totensorflow.keras.applications.NFNetF6
.Who will benefit with this feature?
Anyone doing image classification, including transfer learning and fine tuning; a large portion of the machine learning community!
Any Other info.
DeepMind's implementation Other reference implementations (some incomplete):