вы можете добиться этого путем нарезки индексов.
вот пример для 1D-массива.
a = np.arange(200)
a[100::5] # from index 100 to end with increments of 5
>>> array([100, 105, 110, 115, 120, 125, 130, 135, 140, 145, 150, 155, 160,
165, 170, 175, 180, 185, 190, 195])
для вашего случая
shape = [20,512,512]
arr = np.random.randint(0, 100, shape) # creating an array with the required shape
arr[:, 100::5, 100] # index slicing
>>> array([[88, 74, 45, ..., 72, 33, 63],
[88, 26, 53, ..., 47, 78, 16],
[26, 54, 85, ..., 89, 81, 66],
...,
[76, 1, 11, ..., 3, 74, 7],
[93, 34, 84, ..., 84, 73, 79],
[77, 10, 61, ..., 4, 21, 19]])
arr[:, 100::5, 100].shape
>>> (20, 83)