Downloading MNIST dataset with torchvision gives HTTP Error 503

[omarfoq]

🐛 Bug

I am getting a 503 error when I try to download MNIST dataset with torchvision 0.9.0.
This is probably due to the fact that http://yann.lecun.com/exdb/mnist/ is unavailable, so apart from pproviding a new link to the dataset, I don't see how this problem can be solved

To Reproduce

Steps to reproduce the behavior:

from torchvision.datasets import MNIST

MNIST(".", download=True)

cc @pmeier

[baijumeswani]

It appears that http://yann.lecun.com/exdb/mnist/ is unavailable at this time. A temporary solution before downloading MNIST:

from torchvision import datasets
datasets.MNIST.resources = [
            ('https://ossci-datasets.s3.amazonaws.com/mnist/train-images-idx3-ubyte.gz', 'f68b3c2dcbeaaa9fbdd348bbdeb94873'),
            ('https://ossci-datasets.s3.amazonaws.com/mnist/train-labels-idx1-ubyte.gz', 'd53e105ee54ea40749a09fcbcd1e9432'),
            ('https://ossci-datasets.s3.amazonaws.com/mnist/t10k-images-idx3-ubyte.gz', '9fb629c4189551a2d022fa330f9573f3'),
            ('https://ossci-datasets.s3.amazonaws.com/mnist/t10k-labels-idx1-ubyte.gz', 'ec29112dd5afa0611ce80d1b7f02629c')
        ]

[madhavajay]

[pmeier]

This should be fixed in #3544. We will release torchvision==0.9.1 soon to get this in a release

[ghazpar]

So when can we expect the release of 0.9.1 ?

[fmassa]

The release of 0.9.1 is expected to be out in around 10 days

[ghazpar]

Ok, good to know.

But given that the unavailability of the MNIST database will break existing codes, I would have expected an earlier release?

Just my 2c.

[fmassa]

FYI we've released torchvision 0.9.1 a few days ago https://github.com/pytorch/vision/releases/tag/v0.9.1