Custom Dataset __getitem__ is receiving a list from DataLoader

vision
1

Im new using pytorch. I have a custom dataset which is reading images from disk, it looks like this:

class SaffronDataset(Dataset):
    def __init__(self, base_dir, transform=None, target_transform=None):
        self.images_dir = os.path.join(base_dir, 'raw')
        self.labels_dir = os.path.join(base_dir, 'segmentation')
        self.transform = transform
        self.target_transform = target_transform
        self.images = sorted([os.path.join(self.images_dir,x) for x in os.listdir(self.images_dir)])
        self.labels = sorted([os.path.join(self.labels_dir,x) for x in os.listdir(self.labels_dir)])
        

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

    def __getitem__(self, idx):
        images = Image.open(self.images[idx])
        labels = Image.open(self.labels[idx])
        if self.transform:
            images = self.transform(images)
        if self.target_transform:
            labels = self.transform(labels)
        return images, labels

I have a basic DataLoader too:

data_loader = DataLoader(data, batch_size=2, shuffle=True)

When i try to iterate the dataloader, im getting the error TypeError: list indices must be integers or slices, not list. The problem is that getitem() function is receiving a list instead of a single index and i dont understand why, anyone can help me with that?

2

Might you consider printing out idx so you can see what it contains?

3

Idx contains a list with a number of index equals to the bacth_size. For example, in my case with batch_size = 2, idx can take [2,3] value.
The problem is that if I take it into account and deal with it in the getitem() function, returning a list of images. Then I get an error because a single tuple (image, label) is expected as output.

4

Hi, that’s an unusual case. The idx value in __getitem__(self, idx) function should be a single integer value in the range [0,len(dataset)-1] meant for a single sample of the dataset. Getting a list means something is overriding the default behaviour maybe a custom collate function. Do you mind sharing some more code for context and any errors, or print statements?