Fix the PhotoTour download URLs

[ducha-aiki]

This PR fixes #8960 and #8732 issues.
The original author links with PhotoTour datasets are no longer available.

I contribute by creating my mirror and pointing the links to it.

The patch data was downloaded using torchvision 3 years ago, and now repacked and uploaded the CTU in Prague server.

The hashes are different, because I have zipped the data myself with different zip version, as original archives are deleted by the torchvision downloader after unpacking is done.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/9002

• 📄 Preview Python docs built from this PR

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

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

[AntoineSimoulin]

Hey @ducha-aiki, thank you for submitting this PR and attempting to address the issue. In general, we prefer to direct users to official sources and hubs that are maintained by reputable organizations or communities, rather than individual repositories or personal sources. This is because individually maintained sources can potentially pose safety risks due to the lack of transparency, accountability, and traceability. In light of this, I would prefer not to merge this PR in its current state.
Thank you for your understanding, and please let me know if you have any further questions or concerns.
Best,
Antoine

[ducha-aiki]

@AntoineSimoulin I do understand it. However, have you considered that both links that torchvision used before, are the same thing that I just created- personal websites under the university domain. So the policy you are referring to was already violated. Would it help if I upload to HF datasets?
I also would argue that as a known researcher and kornia maintainer, I am a bit more reputable source than random person.

If you still believe that it is unacceptable, then the next best option is to delete this dataset reference altogether .

--
Best, Dmytro Mishkin.

[NicolasHug]

Thanks for the PR @ducha-aiki

We certainly don't question your integrity and reputation. The main concern is more about sourcing the datasets from their official source, or from a vetted organization where we'd get some guarantees of maintainability. The original links we currently serve, for example, are from Matthew Brown, one of the dataset authors.

I just realized that the official dataset page moved to https://phototour.cs.washington.edu/patches/default.htm, where I can find some links to download the dataset. They don't seem to be exactly in the same format (not the same amount of file) as what we declare in our code, but maybe we can simply update the links to those?

[ducha-aiki]

@NicolasHug I understand your concerns. Regarding the dataset - it was not moved. The link you are referencing, is the earlier (worse) version of the dataset, published at CVPR2007 (as referenced on the page).
The disappeared one was an improved version of it - published at PAMI 2010, and all the patch descriptor community (HardNet, L2Net, TFeat, etc), was using the new one.
So I'd say that if you add https://phototour.cs.washington.edu/patches/default.htm, it should come at different name

[ducha-aiki]

@NicolasHug actually, you know what? Let me try to reach the original link owners here and see if we can just respore everything first.