Update random_cutout.py to be a subclass of VectorizedBaseImageAugmentationLayer

[sup3rgiu]

What does this PR do?

As of v0.6.4, RandomCutout is not implemented in a vectorized manner. This can be easily fixed by making it a subclass of VectorizedBaseImageAugmentationLayer and fixing the shape of some computations.

A full working vectorized implementation+benchmark is available at:
https://colab.research.google.com/drive/1luk08YFpR4VFDW_qD1RHgNFaJDV46Dlf?usp=sharing

Execution time over 10 CIFAR-10 batches of size 128:

Original implementation: 0:00:07.374743
Vectorized implementation:   0:00:00.159785

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue? Please add a link
  to it if that's the case.
• Did you write any new necessary tests?
• If this adds a new model, can you run a few training steps on TPU in Colab to ensure that no XLA incompatible OP are used?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[sampathweb]

Thanks for the PR!. Can you rebase with Master

[sup3rgiu]

Thanks for the PR!. Can you rebase with Master

Done

[divyashreepathihalli]

@sup3rgiu can you please add a demo output of this layer and a colab to check the results?

[sup3rgiu]

@sup3rgiu can you please add a demo output of this layer and a colab to check the results?

Sure!

Colab: https://colab.research.google.com/drive/1luk08YFpR4VFDW_qD1RHgNFaJDV46Dlf

Demo output:

The "brightness" difference of the random noise is simply because in the vectorized version I added few instructions to rescale the noise to the input image range. I think this might be useful to apply the noise correctly, regardless of the image range, but if it isn't you can simply remove these lines within the _compute_rectangle_fill method:

image_max = tf.reduce_max(inputs)
image_min = tf.reduce_min(inputs)
fill_max = tf.reduce_max(fill_value)
fill_min = tf.reduce_min(fill_value)
fill_value = (image_max - image_min) * (fill_value - fill_min) / (fill_max - fill_min) + image_min

[divyashreepathihalli]

Great work! Rescaling the noise to the input image range is cool!
However, there is one failing test -keras_cv/layers/preprocessing/ragged_image_test.py::RaggedImageTest::test_preserves_ragged_status_RandomCutout - ValueError: Exception encountered when calling layer 'random_cutout' (type RandomCutout).
Also, you will need to run shell/format.sh to reformat your code.

[sup3rgiu]

I fixed the implementation to support RaggedTensors. The test should be ok now.
I've also run shell/format.sh to reformat the code.

New colab if needed: https://colab.research.google.com/drive/1luk08YFpR4VFDW_qD1RHgNFaJDV46Dlf

[divyashreepathihalli]

/gcbrun

[divyashreepathihalli]

Thank you! triggering the tests again!