Torchvision/fix issue 4295

[vilktor370]

A new example in the gallery is added as requested in this #4295. This new example is to show the difference between to_tensor, pil_to_tensor, and to_pil_image.

cc @vfdev-5

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/7985

• 📄 Preview Python docs built from this PR

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

❌ 1 New Failure, 1 Unrelated Failure

As of commit caee390 with merge base 35677ec ():

NEW FAILURE - The following job has failed:

• python-source-and-configs / linux-job (gh)

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• unittests-macos (3.8, macos-m1-12) / macos-job (gh)

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

[vilktor370]

@NicolasHug

Are there any known issues regarding the CI pipeline? I am not sure why the CI pipeline is failing on Lint. I did ufmt and flake8 locally on that file. Everything seemed to be good to me before I pushed. I guess it's because I added a new entry in conf.py.

[NicolasHug]

Hi @vilktor370 and thank for the PR.

Quite a bit has changed since we opened #4295 and we have new "preferred" ways of performing these conversions. In particular, we have deprecated ToTensor and we have a new ToImage transform now, that can probably be preferred over PilToImage.

With these new transforms, the overall story should be a bit simpler and probably doesn't require a full-fledge gallery example anymore. I think we can write a small paragraph here https://pytorch.org/vision/main/transforms.html#conversion instead, and explain that:

• To convert from anything to a tensor, ToImage() can be used (there is also PilToTensor(), but it doesn't support numpy).
• To convert to a PIL Image, use ToPILImage()
• To convert the dtype and to scale values of a tensor, ToDtype(dtype, scale=True) can be used
• ToTensor() is deprecated and shouldn't be used anymore. Compose([ToImage(), ToDtype(torch.float32, scale=True)]) is recommended instead.

Does this make sense?