def test_gamma_float_equal_uint8():
img = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)
img_f = img.astype(np.float32) / 255.0
gamma = 0.5
img = F.gamma_transform(img, gamma)
img_f = F.gamma_transform(img_f, gamma)
img = img.astype(np.float32)
img_f *= 255.0
assert (np.abs(img - img_f) <= 1).all()
def __getitem__(self, index):
img_path = self.images[index]
img = cv2.cvtColor(cv2.imread(img_path), cv2.COLOR_BGR2RGB)
img = F.gamma_transform(img, GAMMA_CORRECTION)
if self.combination_type == "STAPLE" :
target = cv2.imread(self.staple_target[index], cv2.IMREAD_UNCHANGED)
else :
t1 = cv2.imread(self.target1[index], cv2.IMREAD_UNCHANGED)
t2 = cv2.imread(self.target2[index], cv2.IMREAD_UNCHANGED)
target = self.combine_multiple_targets(t1, t2)
if self.get_disks:
target_disk = cv2.imread(self.disks[index], cv2.IMREAD_UNCHANGED)
if self.transforms is not None:
if self.get_disks:
augmented = self.transforms(image=img, masks=[target, target_disk])
else:
augmented = self.transforms(image=img, mask=target)
img = augmented['image']
# pick only green channel
if self.green_only:
img = img[[1]]
if self.get_disks:
target = torch.from_numpy(augmented['masks'][0]).long() / 255
target_disk = torch.from_numpy(augmented['masks'][1]).long() / 255
else:
target = augmented['mask'].long() / 255
if self.get_disks:
return img, target, target_disk
else:
return img, target
def preprocess_image(img, orig='DRIVE', gray=True):
_, vt_ = get_transforms(orig)
gamma_ = config.GAMMA_CORRECTION
img_t = gamma_transform(img, gamma_)
img_t = vt_(image=img_t)['image']
if gray:
img_t = img_t[[1]].unsqueeze(0) # shape (B, 1, H, W)
return img_t
def __getitem__(self, index):
from interrater.config import GAMMA_CORRECTION
target = self.target[index]
img_path = self.images[index]
try :
img = cv2.cvtColor(cv2.imread(img_path), cv2.COLOR_BGR2RGB)
img = F.gamma_transform(img, GAMMA_CORRECTION)
if self.transforms is not None:
augmented = self.transforms(image=img)
img = augmented['image']
# pick only green channel (even if no transformation is applied)
if self.green_only:
img = img[[1]]
return img, target
except :
pdb.set_trace()
def __getitem__(self, index):
img_path = self.images[index]
mask_path = self.masks[index]
img = cv2.cvtColor(cv2.imread(img_path), cv2.COLOR_BGR2RGB)
img = F.gamma_transform(img, GAMMA_CORRECTION)
mask = imageio.imread(mask_path)
if self.train:
tgt_path = self.targets[index]
target = imageio.imread(tgt_path)
if self.transforms is not None:
augmented = self.transforms(image=img, masks=[mask, target])
img = augmented['image']
if self.green_only:
img = img[[1]]
mask, target = augmented['masks']
# if isinstance(img, np.ndarray):
# img[mask == 0] = 0
# else:
# img[:, mask == 0] = 0
target = target.astype(int) / 255
if self.return_mask:
return img, torch.from_numpy(mask).long(), torch.from_numpy(target).long()
else:
return img, torch.from_numpy(target).long()
else:
if self.transforms is not None:
augmented = self.transforms(image=img, mask=mask)
img = augmented['image']
if self.green_only:
img = img[[1]]
mask = augmented['mask']
# if isinstance(np.ndarray, img):
# img[mask == 0] = 0
# else:
# img[:, mask == 0] = 0
if self.return_mask:
return img, torch.from_numpy(mask).long()
return img
def albumentations(self, img):
return albumentations.gamma_transform(img, gamma=0.5)
def test_gamma_transform_float(gamma, expected):
img = np.ones((100, 100, 3), dtype=np.float32) * 0.4
expected = np.ones((100, 100, 3), dtype=np.float32) * expected
img = F.gamma_transform(img, gamma=gamma)
assert img.dtype == np.dtype("float32")
assert np.allclose(img, expected)
def test_gamma_transform(gamma, expected):
img = np.ones((100, 100, 3), dtype=np.uint8)
img = F.gamma_transform(img, gamma=gamma)
assert img.dtype == np.dtype("uint8")
assert (img == expected).all()
def test_gamma_transform(gamma, expected):
img = np.ones((100, 100, 3), dtype=np.uint8)
img = F.gamma_transform(img, gamma=gamma)
assert img.dtype == np.dtype('uint8')
assert (img == expected).all()
def albumentations(self, img):
return albumentations.gamma_transform(img, gamma=0.5)
