def test_invalid_block_size():
image = cp.arange(4 * 6).reshape(4, 6)
with pytest.raises(ValueError):
block_reduce(image, [1, 2, 3])
with pytest.raises(ValueError):
block_reduce(image, [1, 0.5])
def test_block_reduce_max():
image1 = cp.arange(4 * 6).reshape(4, 6)
out1 = block_reduce(image1, (2, 3), func=cp.max)
# fmt: off
expected1 = cp.array([[8, 11], [20, 23]])
# fmt: on
assert_equal(expected1, out1)
image2 = cp.arange(5 * 8).reshape(5, 8)
out2 = block_reduce(image2, (4, 5), func=cp.max)
# fmt: off
expected2 = cp.array([[28, 31], [36, 39]])
# fmt: on
assert_equal(expected2, out2)
def test_block_reduce_sum():
image1 = cp.arange(4 * 6).reshape(4, 6)
out1 = block_reduce(image1, (2, 3))
# fmt: off
expected1 = cp.array([[24, 42], [96, 114]])
# fmt: on
assert_equal(expected1, out1)
image2 = cp.arange(5 * 8).reshape(5, 8)
out2 = block_reduce(image2, (3, 3))
# fmt: off
expected2 = cp.array([[81, 108, 87], [174, 192, 138]])
# fmt: on
assert_equal(expected2, out2)
def test_block_reduce_min():
image1 = cp.arange(4 * 6).reshape(4, 6)
out1 = block_reduce(image1, (2, 3), func=cp.min)
# fmt: off
expected1 = cp.array([[0, 3], [12, 15]])
# fmt: on
assert_equal(expected1, out1)
image2 = cp.arange(5 * 8).reshape(5, 8)
out2 = block_reduce(image2, (4, 5), func=cp.min)
# fmt: off
expected2 = cp.array([[0, 0], [0, 0]])
# fmt: on
assert_equal(expected2, out2)
def test_block_reduce_mean():
image1 = cp.arange(4 * 6).reshape(4, 6)
out1 = block_reduce(image1, (2, 3), func=cp.mean)
# fmt: off
expected1 = cp.array([[4., 7.], [16., 19.]])
# fmt: on
assert_equal(expected1, out1)
image2 = cp.arange(5 * 8).reshape(5, 8)
out2 = block_reduce(image2, (4, 5), func=cp.mean)
# fmt: off
expected2 = cp.array([[14., 10.8], [8.5, 5.7]])
# fmt: on
assert_equal(expected2, out2)
def test_block_reduce_median():
if not hasattr(cp, "median"):
pytest.skip("cupy.median is not available")
return
image1 = cp.arange(4 * 6).reshape(4, 6)
out1 = block_reduce(image1, (2, 3), func=cp.median)
# fmt: off
expected1 = cp.array([[4., 7.], [16., 19.]])
# fmt: on
assert_equal(expected1, out1)
image2 = cp.arange(5 * 8).reshape(5, 8)
out2 = block_reduce(image2, (4, 5), func=cp.median)
# fmt: off
expected2 = cp.array([[14., 6.5], [0., 0.]])
# fmt: on
assert_equal(expected2, out2)
image3 = cp.array([[1, 5, 5, 5], [5, 5, 5, 1000]])
out3 = block_reduce(image3, (2, 4), func=cp.median)
assert_equal(5, out3)
def test_func_kwargs_same_dtype():
# fmt: off
image = cp.array([[97, 123, 173, 227], [217, 241, 221, 214],
[211, 11, 170, 53], [214, 205, 101, 57]],
dtype=cp.uint8)
# fmt: on
out = block_reduce(image, (2, 2),
func=cp.mean,
func_kwargs={"dtype": cp.uint8})
expected = cp.array([[41, 16], [32, 31]], dtype=cp.uint8)
assert_equal(out, expected)
assert out.dtype == expected.dtype
def test_func_kwargs_different_dtype():
# fmt: off
image = cp.array([[0.45745366, 0.67479345, 0.20949775, 0.3147348],
[0.7209286, 0.88915504, 0.66153409, 0.07919526],
[0.04640037, 0.54008495, 0.34664343, 0.56152301],
[0.58085003, 0.80144708, 0.87844473, 0.29811511]],
dtype=cp.float64)
# fmt: on
out = block_reduce(image, (2, 2),
func=cp.mean,
func_kwargs={"dtype": cp.float16})
expected = cp.array([[0.6855, 0.3164], [0.4922, 0.521]], dtype=cp.float16)
# Note: had to set decimal=3 for float16 to pass here when using CuPy
cp.testing.assert_array_almost_equal(out, expected, decimal=3)
assert out.dtype == expected.dtype