Missing reshape in torchvision

vision
1

I am missing a transformation in the torchvision datasets. I think it would be grate to have a method that could make a batchcolormap28x28 Mnist image (as example) to a 1*784 vector, to train fully connected network or whatever.

1 Like
2

Maybe I do not understand your question. :sweat_smile:

what you want is some operation like np.reshape? You can use view operation.
examples from doc

>>> x = torch.randn(4, 4)
>>> x.size()
torch.Size([4, 4])
>>> y = x.view(16)
>>> y.size()
torch.Size([16])
>>> z = x.view(-1, 8)  # the size -1 is inferred from other dimensions
>>> z.size()
torch.Size([2, 8])
3

You can do that in torch instead of torchvision using torch.view as @paul_c suggested. The call would be x.view(x.size(0), -1)

4

Yes, for torch tensor is pretty clear.

I meant as part of http://pytorch.org/docs/master/torchvision/transforms.html torchvision.transform. So I could for example load mnist using the MNIST loader and automatically iterate over the train_loader such as each element would be batch_size * dimension instead of batch_size * feature_map * height * width.
It seems torchvision package is only provided for convolutional (and it make sense) but I think could be interesting to have this functionality if you want to train a fully connected for whatever reason.

5

The whole torchvision is python code, you can modify or expand it easily~ :wink:
torchvision

1 Like
6

It is my first day with pytorch, I should first learn it :smile:
I found another way. So, for anyone:

train_feat and train_labl are numpy arrays with mnist with the shape I need.

mnist_dataset_train = torch.utils.data.TensorDataset(torch.from_numpy(train_feat).cuda(), torch.from_numpy(train_labl).cuda())

train_loader = torch.utils.data.DataLoader(mnist_dataset_train,batch_size=100,shuffle=True)

and just

for batch in train_loader:

1 Like
7

Here’s ReshapeTransform I wrote:

class ReshapeTransform:
    def __init__(self, new_size):
        self.new_size = new_size

    def __call__(self, img):
        return torch.reshape(img, self.new_size)

You can use this for MNIST like this:

        mnist_transforms=[
            transforms.ToTensor(),
            transforms.Normalize((0.1307,), (0.3081,)),
            mnist_transforms.append(ReshapeTransform((-1,)))]
        train_ds = datasets.MNIST(train=True, download=True, 
            transform=transforms.Compose(mnist_transforms));
1 Like