View() received an invalid combination of arguments - got (tuple), but expected one of: * (torch.dtype dtype)

vision
1

Hello, I’m trying to build a Generator for my GAN and this is what it looks like:

class Generator(nn.Module):
    def __init__(self, latent_size, init_resolution=8, resolution=256, channels=3):
        super(Generator,self).__init__()
        #resolution = self.resolution
        #channels = self.channels

        self.cnn_torch = nn.Sequential()

        self.cnn_torch.add_module("Linear1", nn.Linear(latent_size, 1024, bias=False))
        self.cnn_torch.add_module("ReLu1", nn.ReLU(inplace=True))
        self.cnn_torch.add_module("Linear2", nn.Linear(1024, 128 * init_resolution * init_resolution, bias=False))
        self.cnn_torch.add_module("ReLu2", nn.ReLU(inplace=True))
        self.cnn_torch.add_module("reshape", Reshape((128, init_resolution, init_resolution)))

        crt_res = init_resolution

        #upsample
        while crt_res != resolution:
          self.cnn_torch.add_module("Upsample", nn.UpsamplingNearest2d(scale_factor=2))
          if crt_res < resolution/2:
            self.cnn_torch.add_module("Upsample Conv2d", nn.Conv2d(crt_res, 256, (5, 5), padding=1, bias=False))
            self.cnn_torch.add_module("Upsample ReLU", nn.ReLU(inplace=True))
          else:
            self.cnn_torch.add_module("Upsample Conv2d - else", nn.Conv2d(crt_res, 128, (5, 5), padding=1, bias=False))

          crt_res = crt_res * 2
          assert crt_res <= resolution,\
                  "Error: final resolution [{}] must equal i*2^n. Initial resolution i is [{}]. n must be a natural number.".format(resolution, init_resolution)

        self.cnn_torch.add_module("Conv2d final", nn.Conv2d(crt_res, channels, (2,2), padding=1, bias=False))
        self.cnn_torch.add_module("Tanh", nn.Tanh())
        print(self.cnn_torch)

    def forward(self, latent):
        fake_image_from_latent = self.cnn_torch(latent)      
        return fake_image_from_latent

But I’m getting the following error when simply trying to test it (with no images):

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-30-ccb426d28f8b> in <cell line: 14>()
     12 
     13 # Pass the latent vector through the generator
---> 14 fake_images = generator(latent_vector)
     15 
     16 # Inspect the generated fake images

5 frames
<ipython-input-14-8cb072a2d5c1> in forward(self, x)
      7 
      8     def forward(self, x):
----> 9         return x.view(self.shape)

TypeError: view() received an invalid combination of arguments - got (tuple), but expected one of:
 * (torch.dtype dtype)
      didn't match because some of the arguments have invalid types: (!tuple of (tuple,)!)
 * (tuple of ints size)
      didn't match because some of the arguments have invalid types: (!tuple of (tuple,)!)

I’m trying to test it this way to see if it outputs the correct tensor shape:

# Instantiate the Generator
latent_size = 100
init_resolution = 8
resolution = 256
channels = 3

generator = Generator(latent_size, init_resolution, resolution, channels)

# Create a random input tensor that represents the latent vector
batch_size = 10
latent_vector = torch.randn(batch_size, latent_size)

# Pass the latent vector through the generator
fake_images = generator(latent_vector)

# Inspect the generated fake images
print(fake_images.shape)

Could anyone help, please? Thanks in advance.

2

Could you print(self.shape) and check what exactly it contains?

3

I tried printing self.shape inside my forward definition but it simply says 'Generator' object has no attribute 'shape', my latent vector however is torch.Size([10, 100]) as expected
Also this is what the print(self.cnn_torch) looks like inside the Generator:

Sequential(
  (Linear1): Linear(in_features=100, out_features=1024, bias=False)
  (ReLu1): ReLU(inplace=True)
  (Linear2): Linear(in_features=1024, out_features=8192, bias=False)
  (ReLu2): ReLU(inplace=True)
  (reshape): Reshape()
  (Upsample): UpsamplingNearest2d(scale_factor=2.0, mode='nearest')
  (Upsample Conv2d): Conv2d(64, 256, kernel_size=(5, 5), stride=(1, 1), padding=(1, 1), bias=False)
  (Upsample ReLU): ReLU(inplace=True)
  (Upsample Conv2d - else): Conv2d(128, 128, kernel_size=(5, 5), stride=(1, 1), padding=(1, 1), bias=False)
  (Conv2d final): Conv2d(256, 3, kernel_size=(2, 2), stride=(1, 1), padding=(1, 1), bias=False)
  (Tanh): Tanh()
)
4

This doesn’t match the previously reported error message since it claims shape is an unexpected tuple while you are not able to print it now as it seems to be undefined.

5

OK, I forgot to include this part of the code which seems to be causing the problem, however I haven’t found a solution yet to have the same behaviour:

class Reshape(nn.Module):
    def __init__(self, *args):
        super(Reshape, self).__init__()
        self.shape = args

    def forward(self, x):
        print(self.shape) #Added this now to check
        return x.view(self.shape)

I was trying to replicate this Keras’ Reshape in PyTorch :
cnn.add(Reshape((128, init_resolution, init_resolution)))
and as you can see I’m using it inside my Generator as the fifth layer (including ReLU layers) and this is the output of print(self.shape):
((128, 8, 8),)

6

Thanks for the code!
Remove the * in *args and it should work.

1 Like
7

Thanks for the help!