RuntimeError: Expected object of scalar type Double but got scalar type Long for sequence element 1 in sequence argument at position #1 'tensors'

1

I am trying to iterate through a dataloader, but I sometimes get a Runtime Error and don’t know how to solve it.

my dataset:

class MyDataset(Dataset):
    
    def __init__(self, label_dir, image_dir, image_names, transform=None):
        self.label_dir = label_dir
        self.image_dir = image_dir
        self.image_names = image_names
        self.transform = transform

    def __len__(self):
        
        return sum(1 for line in open(self.image_names))

    def __getitem__(self, idx):
        img_name = ''
        with open(self.image_names) as file:
            for i, line in enumerate(file):
                if (i == idx):
                    img_name = line[:-1]
                    
        image = io.imread(self.image_dir + img_name + '.jpg')
        label = pd.read_csv(self.label_dir + img_name + '.csv').as_matrix()
        sample = {'image': image, 'label': label}

        if self.transform:
            sample = self.transform(sample)

        return sample

this is how I try to iterate:

dataset = MyDataset(label_dir='labels/',
                                    image_dir='images/', 
                                    image_names = 'names.txt',
                                    transform=transforms.Compose([ToTensor()]))

dataloader = DataLoader(dataset, batch_size=4, shuffle=True)

for i, data in enumerate(dataloader):
    print(i, data['image'].size(), data['label'].size())

When I try to run the for-loop I get the following Error-message:

RuntimeError                              Traceback (most recent call last)
<ipython-input-246-2202a1a6f738> in <module>
----> 1 for i, data in enumerate(dataloader):
      2     print(i, data['image'].size(), data['label'].size())

~\AppData\Local\conda\conda\envs\mt\lib\site-packages\torch\utils\data\dataloader.py in __next__(self)
    558         if self.num_workers == 0:  # same-process loading
    559             indices = next(self.sample_iter)  # may raise StopIteration
--> 560             batch = self.collate_fn([self.dataset[i] for i in indices])
    561             if self.pin_memory:
    562                 batch = _utils.pin_memory.pin_memory_batch(batch)

~\AppData\Local\conda\conda\envs\mt\lib\site-packages\torch\utils\data\_utils\collate.py in default_collate(batch)
     61         return batch
     62     elif isinstance(batch[0], container_abcs.Mapping):
---> 63         return {key: default_collate([d[key] for d in batch]) for key in batch[0]}
     64     elif isinstance(batch[0], tuple) and hasattr(batch[0], '_fields'):  # namedtuple
     65         return type(batch[0])(*(default_collate(samples) for samples in zip(*batch)))

~\AppData\Local\conda\conda\envs\mt\lib\site-packages\torch\utils\data\_utils\collate.py in <dictcomp>(.0)
     61         return batch
     62     elif isinstance(batch[0], container_abcs.Mapping):
---> 63         return {key: default_collate([d[key] for d in batch]) for key in batch[0]}
     64     elif isinstance(batch[0], tuple) and hasattr(batch[0], '_fields'):  # namedtuple
     65         return type(batch[0])(*(default_collate(samples) for samples in zip(*batch)))

~\AppData\Local\conda\conda\envs\mt\lib\site-packages\torch\utils\data\_utils\collate.py in default_collate(batch)
     41             storage = batch[0].storage()._new_shared(numel)
     42             out = batch[0].new(storage)
---> 43         return torch.stack(batch, 0, out=out)
     44     elif elem_type.__module__ == 'numpy' and elem_type.__name__ != 'str_' \
     45             and elem_type.__name__ != 'string_':

RuntimeError: Expected object of scalar type Double but got scalar type Long for sequence element 1 in sequence argument at position #1 'tensors'

The Error only appears if I use a Batch-size for my dataloader. I would be grateful for ideas to solve this problem.

2

The problem was caused, because the tensor-type wasn’t defined in the ToTensor() function. After adding target.type(torch.FloatTensor) everything works.

3

Convert your data to float tensors should solve it:
i.e

dtype = torch.cuda.FloatTensor if torch.cuda.is_available() else torch.FloatTensor
for i, data in enumerate(dataloader):
    data['image'] = data['image'].type(dtype)
    data['label'] = data['label'].type(dtype) #Comment this line if it is a classification problem
    print(i, data['image'].size(), data['label'].size())
3 Likes
4

Thanks for the reply. I used a similar solution, but changed the type inside of the ToTensor() function like this:

class ToTensor(object):
    def __call__(self, sample):
        image, label = sample['image'], sample['label']

        image = image.transpose((2, 0, 1))
        return {'image': torch.from_numpy(image).type(torch.FloatTensor),
                'label': torch.from_numpy(label).type(torch.FloatTensor)}

Now everything works fine.

5

Yes, your solution is good and can work on CPU. But Later during training, if you want to move your data to GPU, you should considerate also the above solution.