Dilate a stack of images with feather kernel

feather_dilate.py

def feather_dilate(stack: np.ndarray, radius=20):
    """Feather dilate a stack of images
    
    stack has dimensions: [image_id, x, y]
    The result is similar to dilation but with feather edges"""
    ms = max_steepness
    msq = ms * 2**0.5
    for x in range(1, stack.shape[2]):
        stack[:,:, x] = (np.maximum(stack[:,:, x], stack[:,:, x-1] - ms))
    for x in range(stack.shape[2]-2, -1, -1):
        stack[:,:, x] = np.maximum(stack[:,:, x], stack[:,:, x+1] - ms)
    for y in range(1, stack.shape[1]):
        stack[:,y, 1:-1] = np.max([stack[:,y, 1:-1], stack[:,y-1, 1:-1] - ms, stack[:,y-1, :-2] - msq, stack[:,y-1, 2:] - msq], axis=0)
    for y in range(stack.shape[1]-2, -1, -1):
        stack[:,y, 1:-1] = np.max([stack[:,y, 1:-1], stack[:,y+1, 1:-1] - ms, stack[:,y+1, :-2] - msq, stack[:,y+1, 2:] - msq], axis=0)

# This function is helpful for training neural networks on sparse masks
# for that purpose, you can gradually decrease the radius during training

# note that this function can be found in Blender but almost nowhere else