def __getitem__(self, index: int):
image = self.read_image(self.image_paths[index], self.feature_size,
self.n_slices)
label = self.read_label(self.label_paths[index], self.feature_size,
self.n_slices)
# template_index = np.random.randint(0, len(self))
# template_index = 0
# template = self.read_label(self.label_paths[template_index], self.feature_size, self.n_slices)
if self.augmentation_prob > 0 and self.augmentation_config is not None:
prob = torch.FloatTensor(1).uniform_(0, 1)
if prob.item() >= self.augmentation_prob:
augment(image,
label,
self.augmentation_config,
(self.n_slices, self.feature_size, self.feature_size),
seed=self.seed)
self.save(image, label, index)
if self.is_3d:
image = np.expand_dims(image, 0)
image = torch.from_numpy(image).float()
label = torch.from_numpy(label).float()
template = torch.from_numpy(self.template).float()
# template_image = np.expand_dims(self.template_image, 0)
# template_image = torch.from_numpy(template_image).float()
return (image, template), (label, template)
def __getitem__(self, index: int):
label = self.read_label(self.label_paths[index], self.feature_size,
self.n_slices)
image = self.read_image(self.image_paths[index], self.feature_size,
self.n_slices)
if self.augmentation_prob > 0 and self.augmentation_config is not None:
image, label = augment(
image,
label,
self.augmentation_config,
(self.n_slices, self.feature_size, self.feature_size),
seed=self.seed)
else:
image, label = random_crop(
image, label,
(self.n_slices, self.feature_size, self.feature_size))
self.save(image, label, index)
if self.is_3d:
image = np.expand_dims(image, 0)
# if self.transform is not None:
# image = torch.from_numpy(image).float()
# label = torch.from_numpy(label).float()
# image = self.transform(image)
# label = self.transform(label)
# self.save(image.numpy(), label.numpy(), index)
image = torch.from_numpy(image).float()
label = torch.from_numpy(label).float()
return image, label
def test(self, index):
label = self.read_label(self.label_paths[index], self.feature_size,
self.n_slices)
image = self.read_image(self.image_paths[index], self.feature_size,
self.n_slices)
image, label = augment(
image,
label,
self.augmentation_config,
(self.n_slices, self.feature_size, self.feature_size),
seed=self.seed)
return image, label
def __getitem__(self, index: int):
image = self.read_image(self.image_paths[index], self.feature_size, self.n_slices)
label = self.read_label(self.label_paths[index], self.feature_size, self.n_slices)
if self.augmentation_prob > 0 and self.augmentation_config is not None:
image, label = augment(
image, label, self.augmentation_config,
(self.n_slices, self.feature_size, self.feature_size),
seed=self.seed
)
else:
image, label = central_crop_with_padding(image, label, (self.n_slices, self.feature_size, self.feature_size))
# image, label = soi_crop(image, label, (self.n_slices, self.feature_size, self.feature_size))
# image, label = random_crop(image, label, (self.n_slices, self.feature_size, self.feature_size))
self.save(image, label, index)
if self.is_3d:
image = np.expand_dims(image, 0)
image = torch.from_numpy(image).float()
label = torch.from_numpy(label).float()
return (image, label), label