ReceptiveField Compute

# [filter size, stride, padding]
#Assume the two dimensions are the same
#Each kernel requires the following parameters:
# - k_i: kernel size
# - s_i: stride
# - p_i: padding (if padding is uneven, right padding will higher than left padding; "SAME" option in tensorflow)
# 
#Each layer i requires the following parameters to be fully represented: 
# - n_i: number of feature (data layer has n_1 = imagesize )
# - j_i: distance (projected to image pixel distance) between center of two adjacent features
# - r_i: receptive field of a feature in layer i
# - start_i: position of the first feature's receptive field in layer i (idx start from 0, negative means the center fall into padding)


import math
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], [1, 1, 0]]
layer_names = ['conv1','pool1','conv2','pool2','conv3','conv4','conv5','pool5','fc6-conv', 'fc7-conv']
imsize = 227


def outFromIn(conv, layerIn):
        n_in = layerIn[0]
        j_in = layerIn[1]
        r_in = layerIn[2]
        start_in = layerIn[3]
        k = conv[0]
        s = conv[1]
        p = conv[2]
        
        n_out = math.floor((n_in - k + 2*p)/s) + 1
        actualP = (n_out-1)*s - n_in + k 
        pR = math.ceil(actualP/2)
        pL = math.floor(actualP/2)
        
        j_out = j_in * s
        r_out = r_in + (k - 1)*j_in
        start_out = start_in + ((k-1)/2 - pL)*j_in
        return n_out, j_out, r_out, start_out


def printLayer(layer, layer_name):
        print(layer_name + ":")
        print("\t n features: %s \n \t jump: %s \n \t receptive size: %s \t start: %s " % (layer[0], layer[1], layer[2], layer[3]))


layerInfos = []
if __name__ == '__main__':
#first layer is the data layer (image) with n_0 = image size; j_0 = 1; r_0 = 1; and start_0 = 0.5
        print ("-------Net summary------")
        currentLayer = [imsize, 1, 1, 0.5]
        printLayer(currentLayer, "input image")
        for i in range(len(convnet)):
                currentLayer = outFromIn(convnet[i], currentLayer)
        layerInfos.append(currentLayer)
        printLayer(currentLayer, layer_names[i])
        print ("------------------------")
        layer_name = raw_input ("Layer name where the feature in: ")
        layer_idx = layer_names.index(layer_name)
        idx_x = int(raw_input ("index of the feature in x dimension (from 0)"))
        idx_y = int(raw_input ("index of the feature in y dimension (from 0)"))
        
        n = layerInfos[layer_idx][0]
        j = layerInfos[layer_idx][1]
        r = layerInfos[layer_idx][2]
        start = layerInfos[layer_idx][3]
        assert(idx_x < n)
        assert(idx_y < n)


print ("receptive field: (%s, %s)" % (r, r))
print ("center: (%s, %s)" % (start+idx_x*j, start+idx_y*j))