Add a GrayscaleToRgb transform that can expand channels to 3

[ahmadsharif1]

This is for issue #8186.

GrayscaleToRgb transform can take a grayscale image and expand it to 3 channels, with the same values for R, G and B.

cc @vfdev-5

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/8247

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (1 Unrelated Failure)

As of commit b2daebd with merge base fa82fd3 ():

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• Tests / unittests-linux (3.9, linux.12xlarge, cpu) / linux-job (gh)
  The process '/usr/bin/git' failed with exit code 128

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[NicolasHug]

Thanks for the PR @ahmadsharif1 ! I think there's a small issue where the images are in fact converted to grayscale instead of being preserved as RGB

[NicolasHug]

Thanks a lot @ahmadsharif1 , just one minor comment below but this looks good. Maybe you can undraft the PR?
Now it's time to bikeshed about the name. I'm tempted to just call that ToRGB() instead of GrayScaleToRGB() so that it can be called at the beginning of every pipeline without suggesting that the input has to be grayscale...

[ahmadsharif1]

Thanks a lot @ahmadsharif1 , just one minor comment below but this looks good. Maybe you can undraft the PR? Now it's time to bikeshed about the name. I'm tempted to just call that ToRGB() instead of GrayScaleToRGB() so that it can be called at the beginning of every pipeline without suggesting that the input has to be grayscale...

I can change it to ToRGB, but while we are bikeshedding, I do like function names that start with a verb or action. So how about ConvertToRGB or MaybeExpandChannelsToRGB or something equivalent? WDYT?

I am happy to just go with ToRGB as well. Just let me know.

[NicolasHug]

As discussed offline, we should just try to be consistent with the existing transforms here (even though the existing transforms themselves are not completely consistent...).

• We already have the Grayscale transform, so I guess that means we should just have RGB()
• We have the rgb_to_grayscale() functional, so lets' have grayscale_to_rgb() as well. (We do have to_grayscale(), but it's an oddity that only supports PIL images for whatever reason... so we shouldn't build on it)

Also before merging we'll have to add the transform and the functional to the docs here and here. To check for docs rendering you can either check out instructions in the contributing guide, or (probably easier for now) just wait for the CI job to be done and then check from #8247 (comment)

[github-actions]

Hey @ahmadsharif1!

You merged this PR, but no labels were added.
The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py