we consider this layer versus adding sequential 3*3 convolution layers such that we get
same global receptive field (assuming only one filter, stride is 1 and no padding)?
121 and 54
11 and 54
0
121 and 45
O
11 and 45
Fig: 1