Computing the network output and the size of the convolution in the padding size explanation TCN

Convolution network output size

Consider only one dimension on the size, the size of a convolutional neural network input is set n, kernel_size is k, padding is p, stride of s, dilation of d, the output size is n '.

Then when d = 1, when S =. 1, ;

To extend the above formula generalized:

Only if d = 1, then ;

If only s = 1, then ;

 

Time convolution in padding the size of the network TCN

TCN only limitation s = 1, so that the normal size of the convolution .

But TCN can only look forward, not look back, so there will be a function of a convolution after TCN behind chomp1d will expand the size of the multi-length padding to delete, therefore, chomp1d after output size .

Since the layer required TCN equal input and output lengths, it should have , so to obtain more comprehensive .