Calculate receptive field and strides for multiple convolutions

receptive_field.py

# https://medium.com/@nikasa1889/a-guide-to-receptive-field-arithmetic-for-convolutional-neural-networks-e0f514068807#1780
import math
def receptive_field(n, layers):
    """
    Parameters
    ----------
    n: int, input size
    layers: array of triplets: (k,s,p)
        k: kernel size
        s: stride
        p: padding
    
    Returns
    -------
    n: output size
    j: jump size
    r: size of receptive field
    start: center position of the receptive field of the first output feature
    """
    r = 1
    j = 1
    start = 0.5
    for k, s, p in layers:
        n = math.floor( (n - k + p)/s ) + 1
        r = r + (k-1)*j
        start = start + ( (k-1)/2 - p)*j
        j = j*s

    return int(n), j, r, start

# Example:
layers = [
    (32, 4, 28), # k, s, p
    (32, 4, 28),
    (1, 4, 0), # maxpool
    (16, 8, 8),
    (16, 8, 8),
]
receptive_field(1024, layers) 
#> (0, 4096, 8796, -640.5)

In line 24 you have p instead of 2*p, which implies that you are using the total padding (i.e. the sum of the paddings applied to each of the borders of the same axis); however, line 26 suggest that you are using the "padding left" value. In particular this function is not exactly equivalent to the one you mention in the Medium article. Which one is right? @svecon