Getting a runtime error for Pytorch trying to implement a neural style transfer learning

I am following this tutorial here. When I try to run the code to get the output image I get this error:

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1, 128, 385, 256]], which is output 0 of AddBackward0, is at version 2; expected version 1 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

I am not sure what I am doing wrong or what to do to correct it. Here is my code:

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from PIL import Image
import matplotlib.pyplot as plt
import torchvision.transforms as transforms 
import torchvision.models as models
import copy
import numpy as np


# In[35]:


# This detects if cuda is available for GPU training otherwise will use CPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)


# In[36]:


# Desired size of the output image
imsize = 512 if torch.cuda.is_available() else 256
print(imsize)


# In[37]:


# Helper function
def image_loader(image_name, imsize):
    # Scale the imported image and transform it into a torch tensor
    loader = transforms.Compose([transforms.Resize(imsize), transforms.ToTensor()])
    image = Image.open(image_name)
    # Fake batch dimension required to fit network's input dimension
    image = loader(image).unsqueeze(0)
    return image.to(device, torch.float)

# Helper function to show the tensor as a PIL image
def imshow(tensor, title=None):
    unloader = transforms.ToPILImage()
    image = tensor.cpu().clone()
    image = unloader(image)
    plt.imshow(image)
    if title is not None:
        plt.title(title)
    plt.pause(0.001) # Pause so that the plots are updated


# In[38]:


# Loading of images
image_directory = './images/'
style_img = image_loader(image_directory + "pb.jpg", imsize)
content_img = image_loader(image_directory + "content.jpg", imsize)
assert style_img.size() == content_img.size(), "we need to import style and content images of the same size"


# In[39]:


plt.ion()


# In[40]:


plt.figure()
imshow(style_img, title='style image')


# In[32]:


plt.figure()
imshow(content_img, title='content image')


# In[58]:


class ContentLoss(nn.Module):
    def __init__(self, target,):
        super(ContentLoss, self).__init__()
        # We detach the target content from the tree used to dynamically
        # compute the gradient: this is stated value,
        # not a variable. Otherwise the forward method of the criterion will throw an error
        self.target = target.detach()
        
    def forward(self, input):
        self.loss = F.mse_loss(input, self.target)
        return input
    
# This is for the syle loss
def gram_matrix(input):
    a, b, c, d = input.size()
    features = input.view(a*b, c*d)
    G = torch.mm(features, features.t()) # compute the gram product
    
    # We normalize the values of the gram matrix by dividing by the number of element in 
    # each feature maps
    return G.div(a*b*c*d)

class StyleLoss(nn.Module):
    def __init__(self, target_feature):
        super(StyleLoss, self).__init__()
        self.target = gram_matrix(target_feature).detach()
        
    def forward(self, input):
        G = gram_matrix(input)
        self.loss = F.mse_loss(G, self.target)
        return input


# In[42]:


# Importing the VGG 19 model like iun the paper (here we set it to evaluation mode)
cnn = models.vgg19(pretrained=True).features.to(device).eval()


# In[43]:


# VGG netowrk are normalized with special values for the mean and std
cnn_normalization_mean = torch.tensor([0.485, 0.456, 0.406]).to(device)
cnn_normalization_std = torch.tensor([0.229, 0.224, 0.225]).to(device)


# In[44]:


# Create a module to normalize the input image so we can easily put it ina nn.Sequential
class Normalization(nn.Module):
    def __init__(self, mean, std):
        super(Normalization, self).__init__()
        # view the mean and std to amke them [C * 1 * 1] so that they can
        # directly work with image Tensor of shape [B * C * H * W]
        # B is a batch size, C is number of channels, H is height, and W is width
        self.mean = torch.tensor(mean).view(-1, 1, 1)
        self.std = torch.tensor(std).view(-1, 1, 1)
        
    def forward(self, img):
        return (img - self.mean) / self.std


# In[46]:


# Desired depth layers to compute style/content losses
content_layers_default = ['conv_4']
style_layers_default = ['conv_1', 'conv_2', 'conv_3', 'conv_5']
num_steps = 300


# In[49]:


def get_style_model_and_losses(cnn, normalization_mean, normalization_std,
                              style_img, content_img,
                              content_layers=content_layers_default,
                              style_layers=style_layers_default):
    cnn = copy.deepcopy(cnn)
    
    # Normalization module
    normalization = Normalization(normalization_mean, normalization_std).to(device)
    
    # Just in order to have an iterable access to or list of content/style losses
    content_losses = []
    style_losses = []
    
    # Assuming that cnn is a nn.Sequential(), so we make a new nn.Sequential
    # to put in modules that are supposed to be activated sequentially
    model = nn.Sequential(normalization)
    
    i = 0
    for layer in cnn.children():
        if isinstance(layer, nn.Conv2d):
            i += 1
            name = 'conv_{}'.format(i)
        elif isinstance(layer, nn.ReLU):
            name = 'relu_{}'.format(i)
            layer = nn.ReLU(inplace=False)
        elif isinstance(layer, nn.MaxPool2d):
            name = 'pool_{}'.format(i)
        elif isinstance(layer, nn.BatchNorm2d):
            name = 'bn_{}'.format(i)
        else:
            raise RuntimeError('Unrecognized layer: {}'.format(layer.__class__.__name__))
            
        model.add_module(name, layer)
        
        if name in content_layers:
            target = model(content_img).detach()
            content_loss = ContentLoss(target)
            model.add_module("content_loss_{}".format(i), content_loss)
            content_losses.append(content_loss)
            
        if name in style_layers:
            target_feature = model(style_img).detach()
            style_loss = StyleLoss(target_feature)
            model.add_module("style_loss_{}".format(i), style_loss)
            style_losses.append(style_loss)
            
            
    # Now we trum off the layers after the last content and style losses
    for i in range(len(model) - 1, -1, -1):
        if isinstance(model[i], ContentLoss) or isinstance(model[i], StyleLoss):
            break
            
    model = model[:(i+1)]
    
    return model, style_losses, content_losses  


# In[60]:


input_img = content_img.clone()
# input_img = torch.randn(content_img.data.size(),device=device)
plt.figure()
imshow(input_img, title= "Input image")


# In[61]:


def get_input_optimizer(input_img):
    optimizer = optim.LBFGS([input_img.requires_grad_()])
    return optimizer


# In[62]:


def run_style_transfer(cnn, normalization_mean, normalization_std,
                       content_img, style_img, input_img, num_steps=300,
                       style_weight=1000000, content_weight=1):
    print('Building the style transfer model...')
    model, style_losses, content_losses = get_style_model_and_losses(cnn,
                                        normalization_mean, normalization_std, style_img, content_img)
    optimizer = get_input_optimizer(input_img)
    
    print('Optimizing...')
    run = [0]
    while run[0] <= num_steps:
        def closure():
            input_img.data.clamp_(0,1)
            
            optimizer.zero_grad()
            model(input_img)
            
            style_score = 0
            content_score = 0
            
            for style_layer in style_losses:
                style_score += (1/5)*style_layer.loss
                
            for content_layer in content_losses:
                content_score += content_layer.loss
                
            style_score *= style_weight
            content_score *= content_weight
            
            loss = style_score + content_score
            loss.backward()
            
            run[0] += 1
            if run[0] % 50 == 0:
                print("run {}:".format(run))
                print('Style Loss : {:4f} Conent Loss: {:4f}'.format(style_score.item(), content_score.item()))
                print()
                
            return style_score + content_score
        
        optimizer.step(closure)
        
    input_img.data.clamp_(0,1)
    
    return input_img


# In[63]:


output = run_style_transfer(cnn,cnn_normalization_mean, cnn_normalization_std,
                           content_img, style_img, input_img, num_steps=num_steps)
plt.figure()
imshow(output, title='Output image')

Replace the inplace operations such as style_score *= style_weight with their out-of-place versions and rerun the code.

1 Like