Append two datasets (MNIST and SVHN)

1

I have two datasets, MNIST and SVHN.
It is known that the num. test samples of MNIST is 1000 and the num. test samples of SVHN is 26032.
Now, i want to append the test set of SVHN to MNIST, in the sense that test_set_append[9999] is the last element of MNIST, and test_set_append[10000] is the first element of SVHN.
I have already tried with the ConcatDataset:

class ConcatDataset(torch.utils.data.Dataset):
    def __init__(self, *datasets):
        self.datasets = datasets

    def __getitem__(self, i):
        return tuple(d[i] for d in self.datasets)

    def __len__(self):
        return min(len(d) for d in self.datasets)

But the result is not what I want:

test_set = ConcatDataset(test_set_a, test_set_b)
test_set[1]
((tensor([[[0., 0., 0.,  ..., 0., 0., 0.],
           [0., 0., 0.,  ..., 0., 0., 0.],
           [0., 0., 0.,  ..., 0., 0., 0.],
           ...,
           [0., 0., 0.,  ..., 0., 0., 0.],
           [0., 0., 0.,  ..., 0., 0., 0.],
           [0., 0., 0.,  ..., 0., 0., 0.]]]), 2),
 (tensor([[[0.5451, 0.5451, 0.5451,  ..., 0.5176, 0.5216, 0.5255],
           [0.5725, 0.5686, 0.5686,  ..., 0.5373, 0.5333, 0.5333],
           [0.6039, 0.6000, 0.6000,  ..., 0.5686, 0.5569, 0.5490],
           ...,
           [0.6196, 0.6000, 0.5804,  ..., 0.3882, 0.4000, 0.4353],
           [0.5843, 0.5529, 0.5216,  ..., 0.3961, 0.4235, 0.4667],
           [0.5569, 0.5216, 0.4824,  ..., 0.4275, 0.4667, 0.5137]]]), 2))
2

In your custom ConcatDataset you are returning a tuple containing samples of both datasets, but I think torch.utils.data.ConcatDataset would directly work.

3

So, you mean this?
test_set=torch.utils.data.ConcatDataset([test_set_a, test_set_b])

4

Yes, I believe this should work, but let us know if that’s not the case.

5

Yes, it worked. Thank you!