add initial chunk of prototype transforms

[pmeier]

This ports the initial chunk of transforms from https://github.com/pmeier/torchvision-datasets-rework/tree/rfc/transforms. You can find a discussion of this in pmeier/torchvision-datasets-rework#1.

Here is a quick demonstration:

import os.path

from torchvision import io, utils
from torchvision.prototype import transforms, features
from torchvision.transforms.functional import to_pil_image


def show(image, bounding_box):
    annotated_image = utils.draw_bounding_boxes(image, bounding_box.convert("xyxy").unsqueeze(0))
    to_pil_image(annotated_image).show()


image = features.Image(io.read_image(os.path.join("test", "assets", "fakedata", "logos", "rgb_pytorch.jpg")))
bounding_box = features.BoundingBox((60, 30, 12, 12), format="cxcywh", image_size=image.shape[-2:])

show(image, bounding_box)

transform = transforms.Compose(
    transforms.Resize(256),
    transforms.HorizontalFlip(),
)

# this would normally come from a dataset, i.e. `for sample in datasets.load("coco")`
sample = dict(foo=image, bar=bounding_box)
transformed_sample = transform(sample)

show(transformed_sample["foo"], transformed_sample["bar"])

Note to reviewers: there are two changes that make the diff look a lot larger than it is:

1. I've renamed torchvision/prototype/transforms/presets.py to _presets.py in order to make it consistent with the other files there. All the changes in torchvision/prototype/models are just fixing the imports.
2. I've moved some functionality from torchvision/prototype/datasets/utils/_internal.py to torchvision/prototype/utils/_internal.py. This also results in a few import changes in other files.

cc @vfdev-5 @datumbox @bjuncek

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit 8788d32 (more details on the Dr. CI page):

---

• 1/1 failures introduced in this PR

---

1 failure not recognized by patterns:

Job	Step	Action
binary_libtorchvision_ops_android	Build	🔁 rerun

---

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

[fmassa]

LGTM, thanks a lot!

Would be good to add tests in a follow-up PR. Also, it would be instructive to implement the basic presets to see how this combines together (without AutoAugment for now for simplicity).

[datumbox]

@pmeier Some of the tests are failing and they are related (see 1, 2)

Marking with "Request changes" to avoid accidental merges. We can merge after fixing the tests.

[datumbox]

Thanks for fixing the import issue. LGTM.

[github-actions]

Hey @pmeier!

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