从二维 numpy 数组中提取二维等大小补丁的索引

Question

I have a numpy array as follows.

a = np.random.rand(5,6)
a
Out[52]: 
array([[0.08649968, 0.24360955, 0.27972609, 0.21566217, 0.00194021,
        0.69750779],
       [0.09327379, 0.7579194 , 0.34634515, 0.78285156, 0.50981823,
        0.17256468],
       [0.7386456 , 0.78608358, 0.80615647, 0.72471626, 0.14825363,
        0.62044455],
       [0.32171325, 0.10889609, 0.56453828, 0.41675939, 0.09400235,
        0.32373844],
       [0.52850344, 0.0783796 , 0.74144658, 0.2363739 , 0.24535204,
        0.9930051 ]])

Then I used the following function to obtain non overlapped patches from the original array a . I used the code from the previously asked question .

def select_random_windows(arr, number_of_windows, window_size):
    # Get sliding windows
    w = view_as_windows(arr,window_size, 3)

    # Store shape info
    m,n =  w.shape[:2]

    # Get random row, col indices for indexing into windows array
    lidx = np.random.choice(m*n,number_of_windows,replace=False)
    r,c = np.unravel_index(lidx,(m,n))
    # If duplicate windows are allowed, use replace=True or np.random.randint

    # Finally index into windows and return output
    return w[r,c]

Then I called select_random_windows to obtain the following two non overlapped patches, each of size 3x3 .

select_random_windows(a, number_of_windows=2, window_size=(3,3))
Out[54]: 
array([[[0.08649968, 0.24360955, 0.27972609],
        [0.09327379, 0.7579194 , 0.34634515],
        [0.7386456 , 0.78608358, 0.80615647]],

       [[0.21566217, 0.00194021, 0.69750779],
        [0.78285156, 0.50981823, 0.17256468],
        [0.72471626, 0.14825363, 0.62044455]]])

Now, how can I get the index of each of the two patches with respect to the main array a . For instance, the first patch should have index of (1x1) and second patch should have index of (1x4) . Is there any way I can extract these center indices of patches with respect to original array a .

Answer 1

You can just use np.where(X == [value1, value2]) to get the index of values