RuntimeError: Given groups=1, weight[64, 3, 3, 3], so expected input[1, 500, 500, 3] to have 3 channels, but got 500 channels instead

1

can someone please help me with this error
this is my CNN

class DnCNN(nn.Module):
    def __init__(self, channels, num_of_layers=17):
        super(DnCNN, self).__init__()
        kernel_size = 3
        padding = 1
        features = 64
        layers = []
        layers.append(nn.Conv2d(in_channels=channels, out_channels=features, kernel_size=kernel_size, padding=padding, bias=False))
    
        layers.append(nn.ReLU(inplace=True))
        for _ in range(num_of_layers-2):
            layers.append(nn.Conv2d(in_channels=features, out_channels=features, kernel_size=kernel_size, padding=padding, bias=False))
            layers.append(nn.BatchNorm2d(features))
            layers.append(nn.ReLU(inplace=True))
        layers.append(nn.Conv2d(in_channels=64, out_channels=3, kernel_size=kernel_size, padding=padding, bias=False))
        self.dncnn = nn.Sequential(*layers)
    def forward(self, x):
        out = self.dncnn(x)
        return out

this is a part of the training

 # Build model
    net = DnCNN(channels=3, num_of_layers=opt.num_of_layers)
    net.apply(weights_init_kaiming)
    criterion = nn.MSELoss(size_average=False)

the training code runs normally and it outputs an error only at the end of the first epoch
saying
RuntimeError: Given groups=1, weight[64, 3, 3, 3], so expected input[1, 500, 500, 3] to have 3 channels, but got 500 channels instead
as you can see i dont have any number of channels set to 500 and still it outputs this error…

2

Based on the error message it seems like you are passing an image tensor in channel-last ordering ([batch_size, w, h, c]), while PyTorch expects channel-first inputs as [batch_size, c, h, w].
Could you check, if all your inputs have this shape?

3

Yes that was the problem and I fixed it by resizing the input, thank you very much for your reply

4

Good to hear it’s working.
However, I’m a bit afraid when you say “resizing the input”, as a view or reshape won’t yield the desired output.
To swap the axes, use permute, if you haven’t already done so.

5

my input’s shape is [500,500,3], and since, as you said pytorch expects [batch_size,c,h,w] I resized the image by using np.resize(img,(3,500,500)) and then I converted it into a tensor by using torch.from_numpy(img) and then I applyed torch.unsqueeze(img,0), will this give the desired output ? or is there a problem?

6

Yeah, that might be problematic.
Have a look at this small example:

x = np.zeros((500, 500, 3))
x[200:300, 200:300, :] = 1.
plt.imshow(x)

x_reshaped = np.resize(x, (3, 500, 500))
print(x_reshaped.shape)
> (3, 500, 500)
print(x_reshaped.sum(1).sum(1))
> [    0. 30000.     0.]

x_transposed = x.transpose(2, 0, 1)
plt.imshow(x_transposed[0])
print(x_transposed.sum(1).sum(1))
> [10000. 10000. 10000.]

As you can see, the reshape operation will put all ones into the second channel, which is wrong.
Use transpose in numpy or permute in pytorch instead.

7

Okay thank you very much for your help