This ones a little esoteric. I'm developing a wave-net inspired model. First time needing dilated convolutions.

My question is which values is the convolution looking at relative to the output position? For example if I have a Conv1D with kernel_size=4, dilation_rate=2 and padding='same' are the inputs split evenly to both sides so relative positions of input data would be (-3,-1,1,3), or justified to one side giving relative positions like (-6,-4,-2,0) or something different? Is there an easy way to control this behavior?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.