卷积层和池化后的维度

Question

I'm testing some things out with the MNIST dataset. images are 28x28 and my x_train has dimensions (60000,28,28,1). I'm using the following code and getting the following error,

        keras.layers.Co`enter code here`nv2D(64,kernel_size = (10,10),activation = "relu",
                            input_shape = (28,28,1)),
        keras.layers.MaxPooling2D(pool_size = (8,8)),

        keras.layers.Conv2D(32,kernel_size = (4,4),activation = "relu"),
        keras.layers.MaxPooling2D(pool_size = (2,2)),

Negative dimension size caused by subtracting 4 from 2 for 'conv2d_47/Conv2D' (op: 'Conv2D') with input shapes: [?,2,2,64], [4,4,64,64]

By my calculations with no padding and a stride of 1, it looks like I should have something with width and height dimensions of 8 by the end. I'm using the formula

W_new = (W1-K+2P)/S where K is the size of the filter,P is the padding, and S is the stride.

Any pointers on where I'm going wrong here would be greatly appreciated.

Thank you

Answer 1

The formula is [W-K+2P/S]+1. I guess after you calculate the correct dimensions after taking into account +1, you should be good. Do let me know if it worked for you.