def test_concats(self):
img1 = np.ones((2, 2), dtype='uint8')
img2 = np.zeros((2, 2), dtype='uint8')
images = np.array([img1, img2])
labels = np.array([[0, 1], [1, 0]])
augm = Augmentation(mixing_coeff=0.5, vertical_concat_prob=1)
res_image, res_label = augm.apply_random_transform(images, labels)
true_image = np.array([[1, 1], [0, 0]], dtype='uint8')
true_label = np.array([0.5, 0.5])
self.assertTrue(np.array_equal(true_image, res_image))
self.assertTrue(np.array_equal(true_label, res_label))
res_image, res_label = augm.horizontal_concat(images,
labels,
mixing_coeff=0.5)
true_image = np.array([[1, 0], [1, 0]], dtype='uint8')
self.assertTrue(np.array_equal(true_image, res_image))
self.assertTrue(np.array_equal(true_label, res_label))
res_image, res_label = augm.mixed_concat(images,
labels,
mixing_coeff=0.5)
true_image = np.array([[1, 0], [0, 1]], dtype='uint8')
self.assertTrue(np.array_equal(true_image, res_image))
self.assertTrue(np.array_equal(true_label, res_label))
def test_mixup(self):
img1 = np.array([[250, 250], [250, 250]], dtype='uint8')
img2 = np.zeros((2, 2), dtype='uint8')
images = np.array([img1, img2])
labels = np.array([[0, 1], [1, 0]])
augm = Augmentation(mixing_coeff=0.5, mixup_prob=1)
res_image, res_label = augm.apply_random_transform(images, labels)
true_image = np.array([[125, 125], [125, 125]], dtype='uint8')
true_label = np.array([0.5, 0.5])
self.assertTrue(np.array_equal(true_image, res_image))
self.assertTrue(np.array_equal(true_label, res_label))
def get_batch_generator(self, batch_size, generator_args={}, **kwargs):
return DataFrameIterator(self.dataframe,
self.dataset_dir,
Augmentation(**generator_args),
x_col=self.x_col,
y_col=self.y_col,
target_size=self.image_size,
batch_size=batch_size,
seed=42,
interpolation='bilinear',
cache=self.cache,
**kwargs)
def get_fewshot_generator(self,
n_way,
k_shot,
query_size=None,
support_generator_args={},
query_generator_args={},
**kwargs):
return FewShotDataFrameIterator(self.dataframe,
self.dataset_dir,
Augmentation(**support_generator_args),
Augmentation(**query_generator_args),
class_index=self.class_index,
n_way=n_way,
k_shot=k_shot,
query_size=query_size,
x_col=self.x_col,
y_col=self.y_col,
target_size=self.image_size,
seed=42,
interpolation='bilinear',
cache=self.cache,
**kwargs)
def test_zero_probs(self):
img = np.random.uniform(0, 256, (50, 50, 3))
img = np.array([img], dtype='uint8')
augm = Augmentation()
res_image, res_label = augm.apply_random_transform(img)
self.assertTrue(np.array_equal(img, res_image))
def test_crop(self):
img = np.random.uniform(0, 256, (50, 50, 3))
img = np.array([img], dtype='uint8')
augm = Augmentation(crop_prob=1, crop_size=10)
res_image, res_label = augm.apply_random_transform(img)
self.assertEqual(res_image.shape, (1, 10, 10, 3))