Pytorch tutorials for Neural Style transfert

With the code cloned to local machine, removed .cuda() and .cpu(), this error is encountered with Python3.6 + Pytorch 0.1:

Traceback (most recent call last):
  File "Neural_Style.py", line 205, in <module>
    style_score += sl.backward()
  File "Neural_Style.py", line 105, in backward
    self.loss.backward(retain_variables=retain_variables)
  File "/usr/local/lib/python3.6/site-packages/torch/autograd/variable.py", line 146, in backward
    self._execution_engine.run_backward((self,), (gradient,), retain_variables)
  File "/usr/local/lib/python3.6/site-packages/torch/nn/_functions/conv.py", line 48, in backward
    if self.needs_input_grad[0] else None)
  File "/usr/local/lib/python3.6/site-packages/torch/nn/_functions/conv.py", line 119, in _grad_input
    return self._thnn('grad_input', input, weight, grad_output)
  File "/usr/local/lib/python3.6/site-packages/torch/nn/_functions/conv.py", line 161, in _thnn
    return impl[fn_name](self, self._bufs[0], input, weight, *args)
  File "/usr/local/lib/python3.6/site-packages/torch/nn/_functions/conv.py", line 251, in call_grad_input
    grad_input, weight, *args)
RuntimeError: Need gradOutput of dimension 4 and gradOutput.size[1] == 64 but got gradOutput to be of shape: [64 x 2401] at /Users/soumith/code/pytorch-builder/wheel/pytorch-src/torch/lib/THNN/generic/SpatialConvolutionMM.c:50

code:

run = [0]
while run[0] <= 300:
    
    def closure():
        optimizer.zero_grad()
        model.forward(input)
        style_score = 0
        content_score = 0

        for sl in style_losses:
            style_score += sl.backward()
        for cl in content_losses:
            content_score += cl.backward()

        run[0]+=1
        if run[0] % 10 == 0:
            print("run " + str(run) + ":")
            print(style_score.data[0])
            print(content_score.data[0])

        return content_score+style_score

    optimizer.step(closure)

plt.subplot(224)
imshow(input.data)
plt.show()

error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-22-0659e65a7a4a> in <module>()
     21         return content_score+style_score
     22 
---> 23     optimizer.step(closure)
     24 
     25 plt.subplot(224)

C:\WinP\bd36\buQt5\winp64-3.6.x.1\python-3.6.5.amd64\lib\site-packages\torch\optim\lbfgs.py in step(self, closure)
    101 
    102         # evaluate initial f(x) and df/dx
--> 103         orig_loss = closure()
    104         loss = float(orig_loss)
    105         current_evals = 1

<ipython-input-22-0659e65a7a4a> in closure()
      9 
     10         for sl in style_losses:
---> 11             style_score += sl.backward()
     12         for cl in content_losses:
     13             content_score += cl.backward()

<ipython-input-15-ed2dfef0a270> in backward(self, retain_variables)
     15 
     16     def backward(self, retain_variables=True):
---> 17         self.loss.backward(retain_variables=retain_variables)
     18         return self.loss

TypeError: backward() got an unexpected keyword argument 'retain_variables'
````

Hello, cool work! I tried to use the transforms.Normalize() method instead of designing a Normalization class as you did ,but the loss seems not converging, is it unachievable to usetransforms.Normalize() method in your code?

load_transform = transforms.Compose([
    transforms.Resize(
        image_size),  # notice the resized img width is image_size
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406],std=[0.229, 0.224, 0.225])
])


def img_loader(image_name):
    img = Image.open(image_name)
    img = load_transform(img).unsqueeze(0)  
    return img.to(device, torch.float)


style_img = img_loader("./datasets/images/picasso.jpg")  # 650*650
content_img = img_loader("./datasets/images/dancing.jpg")  # 444*444
assert style_img.size() == content_img.size(
), "The content-image and the style-image is not compatiable in shape"


# Define the function to show the tensor(Caution: we need to change the tensor format to PIL format)
def img_show(tensor, title=None):
    img = tensor.cpu().clone()  # we clone the tensor to not do changes on it
    img = img.squeeze(0)  # CHW format
    img = img.detach().numpy().transpose((1, 2, 0))  # HWC format
    img = img * np.array([0.229, 0.224, 0.225]) + \
          np.array([0.485, 0.456, 0.406])
    img = img.clip(0, 1)
    plt.imshow(img)
    #  function plt.imshow() performs on RGB data of float [0-1] or int [0-255]
    if title != None:
        plt.title(title)
    plt.pause(0.5)


plt.ion()
plt.figure()
img_show(style_img, title="Style Image")
plt.figure()
img_show(content_img, title="Content Image")

Hi, In order to use vgg19, the following normalizing step has to be used: (Values of the input image being between 0 and 1) normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) cf : https://github.com/pytorch/vision#models Why arn't you using here, not needed ? Thx :)

Hello,

I have run the code in your Jupyter notebook but the result image after training does not look like the "style-transfer" image that you'd expect. Did I do anything wrong or is there any way to make the image better? Here is the screenshot: