Someone can tell me what is the 1 in the Cnn arch dimension equation?

W2=(W1−F+2P)/S+1

What is 1?

bias?

Someone can tell me what is the 1 in the Cnn arch dimension equation?

W2=(W1−F+2P)/S+1

What is 1?

bias?

No, it is just 1. The core of the equation is counting “how often does the weight filter (aka kernel) fit onto the input”. Without padding and with stride=1, you get `width_out = (width_in - filter_width) + 1`

. Now you can see where it comes from: If you start with `width_in == filter_width`

, you have 1 possible position, and for each additional pixel column, i.e. one larger `width_in`

, you have an additional position. So the 1 is this anchoring of the counting.

Best regards

Thomas