def test_norm_min_max_transform(self):
trafo = NormMinMax(per_channel=False)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
trafo = NormMinMax(per_channel=True)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
outp = trafo(**self.batch_dict)
self.assertTrue(isclose(outp["data"].min().item(), 0.0, abs_tol=1e-6))
self.assertTrue(isclose(outp["data"].max().item(), 1.0, abs_tol=1e-6))
def test_norm_zero_mean_transform(self):
trafo = NormZeroMeanUnitStd(per_channel=False)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
trafo = NormZeroMeanUnitStd(per_channel=True)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
outp = trafo(**self.batch_dict)
self.assertTrue(isclose(outp["data"].mean().item(), 0.0, abs_tol=1e-6))
self.assertTrue(isclose(outp["data"].std().item(), 1.0, abs_tol=1e-6))
def test_norm_std_transform(self):
mean = self.batch_dict["data"].mean().item()
std = self.batch_dict["data"].std().item()
trafo = NormMeanStd(mean=mean, std=std, per_channel=False)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
trafo = NormMeanStd(mean=mean, std=std, per_channel=True)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
outp = trafo(**self.batch_dict)
self.assertTrue(isclose(outp["data"].mean().item(), 0.0, abs_tol=1e-6))
self.assertTrue(isclose(outp["data"].std().item(), 1.0, abs_tol=1e-6))
def test_random_scale_value(self):
trafo = RandomScaleValue(DiscreteParameter((2, )))
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
outp = trafo(**self.batch_dict)
expected_out = self.batch_dict["data"] * 2.0
self.assertTrue((outp["data"] == expected_out).all())
def test_clamp_transform(self):
trafo = Clamp(0, 1)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
outp = trafo(**self.batch_dict)
self.assertTrue((outp["data"] == torch.ones_like(outp["data"])).all())
def test_mirror_transform(self):
trafo = Mirror((0, 1))
outp = trafo(**self.batch_dict)
self.assertTrue(outp["data"][0, 0].allclose(
torch.tensor([[9, 8, 7], [6, 5, 4], [3, 2, 1]]).float()))
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
def test_gamma_transform_scalar(self):
trafo = GammaCorrection(gamma=2)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
trafo = GammaCorrection(gamma=2)
outp = trafo(**self.batch_dict)
expected_out = self.batch_dict["data"].pow(2)
self.assertTrue((outp["data"] == expected_out).all())
def test_random_scale_value(self):
trafo = RandomScaleValue("random")
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
random.seed(0)
rand_val = random.random()
random.seed(0)
outp = trafo(**self.batch_dict)
expected_out = self.batch_dict["data"] * rand_val
self.assertTrue((outp["data"] == expected_out).all())
self.assertEqual(trafo.random_mode, "random")
def test_rot90_transform(self):
random.seed(0)
trafo = Rot90((0, 1), prob=1, num_rots=(1, ))
outp = trafo(**self.batch_dict)
self.assertTrue((outp["data"][0, 0] == torch.tensor([[3, 6, 9],
[2, 5, 8],
[1, 4,
7]])).all())
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
trafo = Rot90((0, 1), prob=0)
data_orig = self.batch_dict["data"].clone()
outp = trafo(**self.batch_dict)
self.assertTrue((outp["data"] == data_orig).all())
def test_mirror_transform(self):
trafo = Mirror((0, 1), prob=1)
outp = trafo(**self.batch_dict)
self.assertTrue((outp["data"][0, 0] == torch.tensor([[9, 8, 7],
[6, 5, 4],
[3, 2,
1]])).all())
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
trafo = Mirror((0, 1), prob=0)
data_orig = self.batch_dict["data"].clone()
outp = trafo(**self.batch_dict)
self.assertTrue((outp["data"] == data_orig).all())
def test_gaussian_noise_transform(self):
trafo = GaussianNoise(mean=75, std=1)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
self.check_noise_distance(trafo)
def test_expoential_noise_transform(self):
trafo = ExponentialNoise(lambd=0.0001)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
self.check_noise_distance(trafo)
def test_noise_transform(self):
trafo = Noise("normal", mean=75, std=1)
self.assertTrue(chech_data_preservation(trafo, self.batch_dict))
self.check_noise_distance(trafo)
