def __getitem__(self, i):
# get random sample
input, target = sample_labeled_input(self.data, self.labels,
self.input_shape)
# perform augmentation if necessary
if self.transform is not None:
input = self.transform(input)
if self.target_transform is not None and len(target) > 0:
target = self.target_transform(target)
if self.mode == 'image-level':
target = target[target.shape[0] // 2, target.shape[1] // 2,
target.shape[2] // 2]
if self.input_shape[0] > 1: # 3D data
return input[np.newaxis, ...], target
else:
return input, target
else: # image-level labels as default weak labels
target = target[target.shape[0] // 2, target.shape[1] // 2,
target.shape[2] // 2]
if self.input_shape[0] > 1: # 3D data
return input[np.newaxis, ...], target
else:
return input, target
def __getitem__(self, i):
# get random sample
if self.n_samples is not None:
input = self.samples_x[self.iter_in_epoch]
target = self.samples_y[self.iter_in_epoch]
self.iter_in_epoch += 1
if self.iter_in_epoch == self.n_samples:
self.iter_in_epoch = 0
data = list(zip(self.samples_x, self.samples_y))
shuffle(data)
self.samples_x, self.samples_y = zip(*data)
else:
input, target = sample_labeled_input(self.data, self.labels, self.input_shape)
# perform augmentation if necessary
if self.transform is not None:
input = self.transform(input)
if self.target_transform is not None and len(target)>0:
target = self.target_transform(target)
target = int(target.sum()>0)
if self.input_shape[0] > 1: # 3D data
return input[np.newaxis, ...], target
else:
return input, target
def __getitem__(self, i):
# get random sample
input, target = sample_labeled_input(self.data, self.labels, self.input_shape)
# perform augmentation if necessary
if self.transform is not None:
input = self.transform(input)
if self.target_transform is not None and len(target)>0:
target = self.target_transform(target)
if self.input_shape[0] > 1: # 3D data
return input[np.newaxis, ...], target[np.newaxis, ...]
else:
return input, target
def __init__(self, input_shape, len_epoch=1000, preprocess='z', transform=None, target_transform=None, n_samples=None):
super(EPFLPixelTestDataset, self).__init__(os.path.join('../data', 'epfl', 'testing.tif'),
os.path.join('../data', 'epfl', 'testing_groundtruth.tif'),
input_shape,
len_epoch=len_epoch,
preprocess=preprocess,
transform=transform,
target_transform=target_transform)
self.n_samples = n_samples
self.iter_in_epoch = 0
if n_samples is not None:
self.len_epoch = n_samples
# extract samples
self.samples_x = []
self.samples_y = []
for n in range(n_samples):
input, target = sample_labeled_input(self.data, self.labels, self.input_shape)
self.samples_x.append(input)
self.samples_y.append(target)
def __getitem__(self, i):
# get random sample
r = random.randint(0,len(self.input_shapes)-1)
orientation = self.orientations[r]
input_shape = self.input_shapes[r]
input_shapes = (input_shape[orientation[0]],
input_shape[orientation[1]],
input_shape[orientation[2]])
input, target = sample_labeled_input(self.data, self.labels, input_shapes)
input = np.transpose(input, (orientation.index(0), orientation.index(1), orientation.index(2)))
target = np.transpose(target, (orientation.index(0), orientation.index(1), orientation.index(2)))
# perform augmentation if necessary
if self.transform is not None:
input = self.transform(input)
if self.target_transform is not None and len(target)>0:
target = self.target_transform(target)
if input_shape[0] > 1: # 3D data
return input[np.newaxis, ...], target[np.newaxis, ...]
else:
return input, target
def __init__(self, input_shape, len_epoch=1000, preprocess='z', transform=None, target_transform=None, split=0.5, n_samples=None):
super(MEDPixelTestDataset, self).__init__(os.path.join('../data', 'med', 'data.tif'),
os.path.join('../data', 'med', 'labels.tif'),
input_shape,
len_epoch=len_epoch,
preprocess=preprocess,
transform=transform,
target_transform=target_transform)
s = int(split * self.data.shape[2])
self.data = self.data[:, :, s:]
self.labels = self.labels[:, :, s:]
self.n_samples = n_samples
self.iter_in_epoch = 0
if n_samples is not None:
self.len_epoch = n_samples
# extract samples
self.samples_x = []
self.samples_y = []
for n in range(n_samples):
input, target = sample_labeled_input(self.data, self.labels, self.input_shape)
self.samples_x.append(input)
self.samples_y.append(target)
