[CNN] Why is the convolution kernel size of CNN usually odd

Why? CNN The convolution kernel size of is generally odd ？ chart (a)、(b)、 The convolution kernel sizes of are 3、4、5, among (a) And doing padding It shows great advantages when generating input equal size feature map , However (b) Equal size feature map cannot be realized ,3 When ,padding=1 that will do ;5 When ,padding=2 that will do ;