def test_normalize_greater_1_value_error():
img = np.array([
[[50, 127, 250]],
]) # (1,1,3) image
transform = imagenet.Normalize()
with pytest.raises(ValueError):
actual = transform(img)
def test_auto_normalize_unnormalize():
img = np.array([
[[0.2, 0.5, 0.8]],
]) # (1,1,3) image
normalize = imagenet.Normalize()
unnormalize = imagenet.Unnormalize()
actual = unnormalize(normalize(img))
assert isinstance(actual, torch.Tensor)
actual = actual.numpy().transpose(1, 2, 0)
assert np.isclose(img, actual).all()
def test_normalize_tensor():
img = torch.FloatTensor([
[[0.2, 0.5, 0.8]],
]) # (1, 1, 3) image
img = img.permute(2, 0, 1) # (3, H, W)
transform = imagenet.Normalize()
actual = transform(img)
assert actual.shape == (3, 1, 1)
assert np.isclose(actual[0], (0.2 - 0.485) / 0.229)
assert np.isclose(actual[1], (0.5 - 0.456) / 0.224)
assert np.isclose(actual[2], (0.8 - 0.406) / 0.225)
def test_normalize_uint8_value_error():
img = np.array(
[
[[0, 0, 0]],
],
dtype=np.uint8,
) # (1,1,3) image
transform = imagenet.Normalize()
with pytest.raises(ValueError):
actual = transform(img)
def test_normalize_numpy():
img = np.array([
[[0.2, 0.5, 0.8]],
]) # (1,1,3) image
transform = imagenet.Normalize()
actual = transform(img)
assert isinstance(actual, torch.Tensor)
assert actual.shape == (3, 1, 1)
assert np.isclose(actual[0], (0.2 - 0.485) / 0.229)
assert np.isclose(actual[1], (0.5 - 0.456) / 0.224)
assert np.isclose(actual[2], (0.8 - 0.406) / 0.225)