RuntimeError: Expected object of scalar type long int but got scalar type float for sequence element 1

vision
1

Hello all, I am new to Pytorch and machine learning and recently I tried to run https://github.com/pclucas14/pixel-cnn-pp/blob/master/main.py with my own dataset.
My dataset is as following:

    def __init__(self, datapath, train=False, transform=None):
        self.data = []
        self.map_dic = {"-" : 0, "X" : 1, "S" : 2, "?" : 3, "Q" : 4, "E" : 5, "<" : 6, ">" : 7, "[" : 8, "]" : 9, "o" : 10, "B" : 11, "b" : 12}
        for file_name in os.listdir(datapath): 
            with open(base_path + "/" + file_name, 'r') as f:
               res = np.array( list( map( lambda l: [self.map_dic.get(c, c) for c in l.strip()], f.readlines() ) ) )
            self.data.append(res)
        self.transform = transform

    def __getitem__(self, index):
        img = self.data[index]

        img = torch.from_numpy(img)
        img = img.unsqueeze(0)

        # image transform:
        # img = Image.open('dataset_img/{}.png'.format(index)).convert('L')
        # if self.transform is not None:
        #     img = self.transform(img)

        return img,0
    
    def __len__(self):
        return len(self.data)

The code works fine originally when I load the image from file (using the comment part code) and transform with the following

    transforms.Resize(size=(16,224)),
    transforms.ToTensor()
    ])

The image file here is created bymatplotlib.pyplot.imsave('name1.png', data, cmap='gray') function from the numpy array(like this
[[0 0 0 … 0 0 0]
[0 0 0 … 0 0 0]
[0 0 0 … 0 0 0]
…
[1 1 1 … 1 1 1]
[1 1 1 … 1 1 1]
[1 1 1 … 1 1 1]]
)
Then I think that it is not necessary to convert numpy array to image and then to tensor. Instead I could convert directly from numpy array to tensor. After I use the above dataset, it constantly gives the error:

  File "maintxt1.py", line 241, in <module>
#in original code it is line 122
    output = model(input)
  File "/home/tangyeping/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/tangyeping/pixel-cnn-pp/model.py", line 119, in forward
    x = x if sample else torch.cat((x, self.init_padding), 1)
RuntimeError: Expected object of scalar type long int but got scalar type float for sequence element 1.

I have tried something like img = torch.from_numpy(img).long() or model.double() but this error keeps appering. Appreciate for any suggestion.

2

Firstly you have not applied the defined transformation.
apply the transformation.

second, if you don’t want to apply the transformation,
change the tensor type using the torch.from_numpy(img).float

show output if you still got any errors.

3

Thank you a lot! I did not use transform because now I am dealing with array and tensor instead of image. After I use torch.from_numpy(img).float() this error is going. However another error coming up:

loss : 6.1917, time : 0.4663
Traceback (most recent call last):
  File "maintxt1.py", line 240, in <module>
    output = model(input)
  File "/home/tangyeping/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/tangyeping/pixel-cnn-pp/model.py", line 119, in forward
    x = x if sample else torch.cat((x, self.init_padding), 1)
RuntimeError: Sizes of tensors must match except in dimension 3. Got 184 and 148

I have met this error before when using image object. And after I apply transforms.Resize(size=(16,224)) the error is gone. Is there anything similar I can do with tensor or numpy array to fix it? Or is there another source of error?

4

Okie.

Now as you have not used ToTensor().
you have to reshape the data before feeding to model.
input = input.permute(batch_size, input_channel, n,m)

5

Thank you for the advice. I thought what permute can do is switching the dimension of tensor. I try to print the size of input and I got

starting training
torch.Size([1, 1, 16, 184])
loss : 6.1917, time : 0.4552
torch.Size([1, 1, 16, 148])
Traceback (most recent call last):
  File "maintxt1.py", line 241, in <module>
    output = model(input)
  File "/home/tangyeping/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/tangyeping/pixel-cnn-pp/model.py", line 119, in forward
    x = x if sample else torch.cat((x, self.init_padding), 1)
RuntimeError: Sizes of tensors must match except in dimension 3. Got 184 and 148

I am not sure how I can use permute to reshape the data like what transform.Resize() does.

6

Cant see 184 in the input size, from where this came from.
The issue is with the size, please check forward function also .
Something is happening here
x = x if sample else torch.cat((x, self.init_padding), 1)

print the sizes in forward function it will help to debug.