def test_elastic_transform_interpolation(monkeypatch, interpolation):
image = np.random.randint(low=0,
high=256,
size=(100, 100, 3),
dtype=np.uint8)
mask = np.random.randint(low=0, high=2, size=(100, 100), dtype=np.uint8)
monkeypatch.setattr(
'albumentations.augmentations.transforms.ElasticTransform.get_params',
lambda *_: {'random_state': 1111})
aug = A.ElasticTransform(alpha=1,
sigma=50,
alpha_affine=50,
interpolation=interpolation,
p=1)
data = aug(image=image, mask=mask)
expected_image = F.elastic_transform(
image,
alpha=1,
sigma=50,
alpha_affine=50,
interpolation=interpolation,
border_mode=cv2.BORDER_REFLECT_101,
random_state=np.random.RandomState(1111))
expected_mask = F.elastic_transform(
mask,
alpha=1,
sigma=50,
alpha_affine=50,
interpolation=cv2.INTER_NEAREST,
border_mode=cv2.BORDER_REFLECT_101,
random_state=np.random.RandomState(1111))
assert np.array_equal(data['image'], expected_image)
assert np.array_equal(data['mask'], expected_mask)
def create_cotton_wool_mask(size=128):
iterations = 25
num_circles = 10
acc = np.zeros((size, size), dtype=np.float32)
for i in range(iterations):
img = np.zeros_like(acc)
for j in range(num_circles):
r = int(random.uniform(5, 25))
r2 = int(random.uniform(5, 25))
a = random.uniform(0, 180)
x = int(random.gauss(size // 2, size // 8))
y = int(random.gauss(size // 2, size // 8))
pt = int(x), int(y)
cv2.ellipse(img,
pt, (r, r2),
a,
0,
360,
color=1,
thickness=cv2.FILLED)
# cv2.circle(img, pt, r, color=1, thickness=cv2.FILLED)
acc += img
acc /= iterations
acc = (acc * 255).astype(np.uint8)
acc_blur = cv2.GaussianBlur(acc,
ksize=(9, 9),
sigmaX=5,
borderType=cv2.BORDER_CONSTANT)
acc_sharp = cv2.addWeighted(acc, 1.2, acc_blur, -0.2, 0, dtype=cv2.CV_8U)
acc = acc_sharp
acc = elastic_transform(acc,
alpha=15,
sigma=12,
alpha_affine=50,
border_mode=cv2.BORDER_CONSTANT)
# cv2.imshow('acc', acc)
# cv2.waitKey(30)
return acc
def create_microaneurisms(image,
center=(256, 256),
radius=140,
color=(100, 255, 0),
num=5,
aneurism_radius=(1, 3),
alpha=0.2):
# mask = image.copy()
aneurism_mask = np.zeros_like(image)
for i in range(num):
x = int(random.gauss(center[0], radius))
y = int(random.gauss(center[0], radius))
x = min(max(x, center[0] - radius), center[1] + radius)
y = min(max(y, center[1] - radius), center[1] + radius)
r = int(random.uniform(aneurism_radius[0], aneurism_radius[1]))
cv2.circle(aneurism_mask, (x, y),
r,
color,
thickness=cv2.FILLED,
lineType=cv2.LINE_AA)
# cv2.circle(mask, (x, y), r, (0, 0, 0), thickness=cv2.FILLED, lineType=cv2.LINE_AA)
aneurism_mask = elastic_transform(aneurism_mask,
alpha=5,
sigma=2,
alpha_affine=0)
cv2.GaussianBlur(aneurism_mask, ksize=(5, 5), sigmaX=0, dst=aneurism_mask)
aneurism_mask = 1.0 - aneurism_mask / 255.
overlay = cv2.addWeighted(image, (1 - alpha),
image * aneurism_mask,
alpha,
0,
dtype=cv2.CV_8U)
return overlay
def __call__(self, sample):
# print('grid', np.random.uniform(0.0, 1.0))
if np.random.uniform(0.0, 1.0) < self.p:
return sample
h, w, _ = sample["img"].shape
# grid_distortion
if np.random.uniform(0.0, 1.0) < self.p:
num_steps=15
distort_limit=[-0.05,0.05]
stepsx = [1 + random.uniform(distort_limit[0], distort_limit[1]) for i in
range(num_steps + 1)]
stepsy = [1 + random.uniform(distort_limit[0], distort_limit[1]) for i in
range(num_steps + 1)]
sample["img"]=albumentations.grid_distortion(sample["img"],5,stepsx, stepsy)
# elastic_transform
else:
sample["img"]=albumentations.elastic_transform(sample["img"], alpha=5, sigma=1, alpha_affine=random.uniform(0,2),
interpolation=cv2.INTER_LINEAR, border_mode=cv2.BORDER_REFLECT_101,)
if np.random.uniform(0.0, 1.0) < self.p-0.2:
sample["img"]=albumentations.jpeg_compression(sample["img"], random.randint(20, 100))
return sample
