RGB to Grayscale

[ghosalsattam]

See #1333 #1353 #1226
This PR is responsible for RGB to Grayscale Conversion of image(s).

[ghosalsattam]

@bhack @gabrieldemarmiesse @seanpmorgan @WindQAQ Please provide a review.

[WindQAQ]

Personally, I don't think we should have this as it is merely a single-lined wrapper of tf.image.rgb_to_grayscale.

tf.image.rgb_to_grayscale has the ability to convert image with shape [H, W, 3] and [B, H, W, 3]. For the 2D image, the function in this PR is going to convert [H, W] to [1, H, W, 1] and raise the exception. It turns out this that function can only do the same functionality as tf.image.rgb_to_grayscale does.

https://colab.research.google.com/drive/1v7C-DkHYJijgIdrwKhMpXmVqrJbLW8Sf

@tensorflow/sig-addons-maintainers Hi all, any thought on this? Thanks.

[gabrieldemarmiesse]

I agree with @WindQAQ , there is not enough added value here compared to what tensorflow offers, we can use directly tf.image.rgb_to_grayscale . I think we need to edit #1353 (comment) . I can understand why someone might think that we need to copy the function here. I'll correct the message.

Thank you @ghosalsattam for the pull request, and sorry for not looking too much into the issue. I'll check the box for rgb to grayscale.

[bhack]

I agree cause in our cause just for this single line call we need +137 lines of diff (overhead, tests etc.) also if the original wrapper was just 5 lines.