size of receptive field of each layer in CNN

[李媛媛]

i have searched an answer on internet, but i don't understand it, it is as follows:

 #Compute input size that leads to a 1x1 output size, among other things  

 

# [filter size, stride, padding]

 

convnet =[[11,4,0],[3,2,0],[5,1,2],[3,2,0],[3,1,1],[3,1,1],[3,1,1],[3,2,0],[6,1,0]]

layer_name = ['conv1','pool1','conv2','pool2','conv3','conv4','conv5','pool5','fc6-conv']

imsize = 227

 

def outFromIn(isz, layernum = 9, net = convnet):

    if layernum>len(net): layernum=len(net)

 

    totstride = 1

    insize = isz

    #for layerparams in net:

    for layer in range(layernum):

        fsize, stride, pad = net[layer]

        outsize = (insize - fsize + 2*pad) / stride + 1

        insize = outsize

        totstride = totstride * stride

    return outsize, totstride

 

def inFromOut( layernum = 9, net = convnet):

    if layernum>len(net): layernum=len(net)

    outsize = 1

    #for layerparams in net:

    for layer in reversed(range(layernum)):

        fsize, stride, pad = net[layer]

        outsize = ((outsize -1)* stride) + fsize

    RFsize = outsize

    return RFsize

 

if __name__ == '__main__':

 

    print "layer output sizes given image = %dx%d" % (imsize, imsize)

    for i in range(len(convnet)):

        p = outFromIn(imsize,i+1)

        rf = inFromOut(i+1)

        print "Layer Name = %s, Output size = %3d, Stride = % 3d, RF size = %3d" % (layer_name[i], p[0], p[1], rf)







and the result is:

but i do as it say, i got different result? could anyone show me more details or different explanation? thank you very much!