How to convert dataset of images to tensor?

Question

I've got a dataset of images that looks like this:

array([[[[0.35980392, 0.26078431, 0.14313725],
     [0.38137255, 0.26470588, 0.15196078],
     [0.51960784, 0.3745098 , 0.26176471],
     ...,
     [0.34313725, 0.22352941, 0.15      ],
     [0.30784314, 0.2254902 , 0.15686275],
     [0.28823529, 0.22843137, 0.16862745]],

    [[0.38627451, 0.28235294, 0.16764706],
     [0.45098039, 0.32843137, 0.21666667],
     [0.62254902, 0.47254902, 0.36470588],
     ...,
     [0.34607843, 0.22745098, 0.15490196],
     [0.30686275, 0.2245098 , 0.15588235],
     [0.27843137, 0.21960784, 0.16176471]],

    [[0.41568627, 0.30098039, 0.18431373],
     [0.51862745, 0.38529412, 0.27352941],
     [0.67745098, 0.52058824, 0.40980392],
     ...,
     [0.34901961, 0.22941176, 0.15588235],
     [0.29901961, 0.21666667, 0.14901961],
     [0.26078431, 0.20098039, 0.14313725]],

    ...,

What I need is convert it to tensor so that I could pass it to CNN. I'm trying to do it like that:

from torchvision import transforms as transforms
transform = transforms.Compose([
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ])

How can I apply this transform to my dataset? Thanks for any help.

Answer 1

you probably want to create a dataloader. You will need a class which iterates over your dataset, you can do that like this:

import torch
import torchvision.transforms

class YourDataset(torch.utils.data.Dataset):
    def __init__(self):

       # load your dataset (how every you want, this example has the dataset stored in a json file
        with open(<dataset-path>, "r") as f:
            self.dataset = json.load(f)

    def __getitem__(self, idx):
        sample = self.dataset[idx]
        data, label = sample[0], sample[1]
        
        transform = transforms.Compose([
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ])
        return transform(data), torch.tensor(label)

    def __len__(self):
        return len(self.dataset)

Now you can create a dataloader:

train_set = YourDataset()

train_dataloader = torch.utils.data.DataLoader(
            train_set,
            batch_size=64,
            num_workers=1,
            shuffle=True,
        )

And now you can iterator over the dataloader in your train-loop:

for samples, labels in self.train_set:
    . . . 
    # samples will hold N samples of your dataset where N is the batchsize

If you need more explanation, take a look a pytorchs documentation on this topic.

Answer 2

After being initialized a torchvision transform can be called on a PIL image or torch.Tensor depending on the transform (in the documentation you can find precisely what is expected by a given transform). So assuming your transform pipeline is valid (output type for each transform is compatible with input type of following transform) then you can simply call it: transform(data).

This should work with your data since transforms.ToTensor can accept a numpy.ndarray as input

---

Alternatively, if your dataset is a little more complex than a simple NumPy array. You could implement your own Dataset class which would handle: image fetching in file system, performing relevant transformations, etc... Here's a rough guideline:

class MyDataset(Dataset):
    def __init__(self):
        super(MyDataset, self).__init__()

        # define your data for example a list of images
        self.data = ...

        # define your transform pipline
        transform = transforms.Compose([
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ])

    def __getitem__(self, index):
        path = self.data[index]
        img = Image.open(path)
        return self.transform(x)

Where Image is imported from PIL.

---

For you case, though, you would have something a little simpler:

class MyDataset(Dataset):
    def __init__(self, data):
        super(MyDataset, self).__init__()

        self.data = data

        # define your transform pipline
        transform = transforms.Compose([
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ])

    def __getitem__(self, index):
        x = self.data[index]
        return self.transform(x)

And you would pass your numpy.ndarray to the MyDataset's initializer.