Migrating Nnclr to Keras-3[TF BackEnd]

[aditya02shah]

This PR migrates the nnclr example to Keras 3.0 as requested in keras-team/keras-cv#2211.

1. The tutorial has tensorflow dependency for data preprocessing due to tf.data and tf.image.crop_and_resize() function.
2. Tested the code with torch and jax backends and works fine.

[divyashreepathihalli]

Thank you for the PR @aditya02shah !!
My suggestion for tf.image.crop_and_resize() is to use the following two functions in Keras 3

• Cropping3D
• Resize

Everything else looks good!

[aditya02shah]

@divyashreepathihalli, thank you for your feedback! I considered your suggestion and opted to replace tf.image.crop_and_resize() with the keras_cv.RandomCropAndResize layer, as it more directly aligns with the required functionality. I've tested these changes, and everything else remains intact.
Please review at your convenience. Appreciate your time!

[divyashreepathihalli]

LGTM! thank you! please generate the .md and .ipynb files.

[aditya02shah]

@divyashreepathihalli I have explicitly set the TensorFlow backend as the default, as utilizing the JAX and PyTorch backends results in errors such as TypeError:can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first

[fchollet]

This example is expected to be TF-only since it uses the GradientTape.

LGTM, thank you!

[fchollet]

Actually, it turns out that the files for this example were generated using Keras 2. Can you regenerate using Keras 3?

[aditya02shah]

@fchollet I have migrated the pull request here .
There was an issue with some dependencies in my environment but now everything has been fixed.
I sincerely apologize for any inconvenience this may have caused and appreciate your understanding. Thank you for your time and consideration!