CODEWITHSUNDEEP
pythonimage-processingcomputer-visionconv-neural-networkpytorch
user11173832
user11173832

Reputation:

RuntimeError: Given input size: (10x7x7). Calculated output size: (10x0x0). Output size is too small

Trying to train the mnist 28x28x1 images

my model is 

   def __init__(self):
        super(CNN_mnist, self).__init__()
        self.conv = nn.Sequential(
            # 3 x 128 x 128
            nn.Conv2d(1, 32, 3, 1, 1),
            nn.BatchNorm2d(32),
            nn.LeakyReLU(0.2),

            # 32 x 128 x 128
            nn.Conv2d(32, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.LeakyReLU(0.2),

            # 64 x 128 x 128
            nn.MaxPool2d(2, 2),

            # 64 x 64 x 64
            nn.Conv2d(64, 128, 3, 1, 1),
            nn.BatchNorm2d(128),
            nn.LeakyReLU(0.2),

            # 128 x 64 x 64
            nn.Conv2d(128, 256, 3, 1, 1),
            nn.BatchNorm2d(256),
            nn.LeakyReLU(0.2),

            # 256 x 64 x 64
            nn.MaxPool2d(2, 2),

            # 256 x 32 x 32
            nn.Conv2d(256, 10, 3, 1, 1),
            nn.BatchNorm2d(10),
            nn.LeakyReLU(0.2)
        )
        # 256 x 32 x 32
        self.avg_pool = nn.AvgPool2d(32)
        # 256 x 1 x 1
        self.classifier = nn.Linear(10, 10)

    def forward(self, x):
        features = self.conv(x)
        flatten = self.avg_pool(features).view(features.size(0), -1)
        output = self.classifier(flatten)
        return output, features

and I got following error

RuntimeError: Given input size: (10x7x7). Calculated output size: (10x0x0). Output size is too small

not sure what this error mean and where should I fix it?

Upvotes: 0

Views: 4129

Answers (1)

Shai
Shai

Reputation: 114786

Your [avg_pool] layer expects its input size to be (at least) 32x32, as the kernel size defined for this layer is 32.
However, given the size of the input, the feature map this pooling layer gets is only 7x7 in size. This is too small for kernel size of 32.

You should either increase the input size, or define a smaller (e.g., 7) kernel size for the avg_pooling layer.

Upvotes: 3

Related Questions