def normalize_image(img_npy):
    """
    :param img_npy: b, c, h, w
    """
    for b in range(img_npy.shape[0]):
        for c in range(img_npy.shape[1]):
            img_npy[b, c] = (img_npy[b, c] - img_npy[b, c].mean()) / img_npy[b, c].std()
    return img_npy


def normalize_image_to_0_1(img):
    return (img-img.min())/(img.max()-img.min())


def normalize_image_to_m1_1(img):
    return -1 + 2 * (img-img.min())/(img.max()-img.min())