Vectorize RandomBrightness, RandomContrast, RandomHue and RandomColorJitter

[james77777778]

Short Description

As #1392 was merged, I'm willing to contribute for vectorized RandomBrightness, RandomContrast, RandomHue and finally to complete vectorized RandomColorJitter.

RandomColorJitter is widely adopted, so the improvement from vectorization should benefit the community.

Other Information

There is an unsolved conversation at #1392 (comment)
and my questions are:

1. Should we move A/B testings from unit tests into ./benchmarks/random_*.py?
2. ./benchmarks/ might be flooded by these scripts. Should we create a new folder for these?

For 1., take ./benchmarks/vectorized_random_saturation.py as example:

# add old implementation
class OldRandomSaturation(BaseImageAugmentationLayer):
    ...

# add tests
class RandomSaturationTest(tf.test.TestCase):
    def test_correctness_with_tf_adjust_saturation_normalized_range(self):
        ...
        self.assertAllClose(old_output, output, atol=1e-5, rtol=1e-5)

    def test_correctness_with_tf_adjust_saturation_rgb_range(self):
        ...
        self.assertAllClose(old_output, output, atol=1e-3, rtol=1e-5)


if __name__ == "__main__":
    # run benchmarks (get some figures for improvement)
    ...
    # run unit tests
    tf.test.main()

For 2., I think something like ./benchmarks/vectorization_strategy/vectorized_*.py can be a better layout?

[IMvision12]

hey @james77777778 Just saw your last PR and that was wonderful work on vectorizing random saturation, are there any more augmentations that need to be vectorized, I would like to contribute.

Thanks

[james77777778]

Hi @IMvision12
Thanks!

Currently, I'm not targeting any other preprocessing layer. Just want to complete RandomColorJitter in my own interest :)
As #1382 (comment) said, any vectorization should be contribution-welcome.