Add new illustrations of transforms

[NicolasHug]

With #3652 merged, we can now illustrate how each transforms affects an image via a sphinx-gallery example:

https://519065-73328905-gh.circle-artifacts.com/0/docs/auto_examples/plot_transforms.html#sphx-glr-auto-examples-plot-transforms-py

The goal of this issue is to add more transforms to this file: https://github.com/pytorch/vision/blob/master/gallery/plot_transforms.py

Instructions:

• pick you favourite transform that isn't already present in the example (after checking below that no one is already working on it)
• comment below "I'm working on <the transform>" so that others don't pick the same one
• edit the example file with something that properly illustrate the transform, taking inspiration from the already present transforms
• Build the docs following the instructions here, check that everything is fine, then submit a PR.

You can pick more than one transform in the same PR if you feel like it.

cc @vfdev-5

[oke-aditya]

Hi @NicolasHug I can pick a few.
Let's keep track.

In order as they appear in Docs

Torchvision Transforms on PIL Image and Torch Tensor.

• CenterCrop
• ColorJitter
• FiveCrop
• Grayscale
• Pad
• RandomAffine
• RandomApply
• RandomCrop
• RandomGrayscale
• RandomHorizontalFlip
• RandomPerspective
• RandomResizedCrop
• RandomRotation
• RandomVerticalFlip
• Resize
  Scale
• GaussianBlur
• AutoAugment

[oke-aditya]

I will pick CenterCrop and FiveCrop will continue further.

[voldemortX]

@oke-aditya isn't Scale kind of deprecated?

[oke-aditya]

Yes it is. Let us not count it

[voldemortX]

I'd like to work on RandomAffine, RandomPerspective and RandomRotation , do I need to do something like fix random seeds?

[NicolasHug]

do I need to do something like fix random seeds?

I didn't set it for the ColorJitter, the idea being that if a user downloads the notebook they will see that a different transformation is performed each time the cell is executed, which is good because it illustrates that these transforms are random at each call.

We should try to set the parameters of the transforms such that it's always obvious that something happened, i.e. that the 2 images are clearly not the same.

[harishsdev]

Hi all,i would like to work on below transforms

RandomApply
RandomCrop
RandomGrayscale
RandomHorizontalFlip
RandomResizedCrop
RandomVerticalFlip
GaussianBlur