def get_numpy_dataset(original_dataset, input_slice, output_slice, transform):
dataset_numpy = dict()
original_data_slice = get_zero_padded_array_slice(original_dataset['data'], input_slice)
data_slice = np.array(original_data_slice, dtype=np.float32) / (2. ** 8)
if transform:
if 'transform' in original_dataset:
lo, hi = original_dataset['transform']['scale']
data_slice = 0.5 + (data_slice-0.5)*np.random.uniform(low=lo, high=hi)
lo, hi = original_dataset['transform']['shift']
data_slice = data_slice + np.random.uniform(low=lo, high=hi)
else:
print("WARNING: source data doesn't have 'transform' attribute.")
dataset_numpy['data'] = data_slice
# load outputs if desired
if output_slice is not None:
dataset_numpy['components'] = get_zero_padded_array_slice(original_dataset['components'], output_slice)
if 'label' in original_dataset:
label_shape = original_dataset['label'].shape
label_slice = (slice(0, label_shape[0]),) + output_slice
dataset_numpy['label'] = get_zero_padded_array_slice(original_dataset['label'], label_slice)
else:
# compute affinities from components
dataset_numpy['label'] = pygt.malis.seg_to_affgraph(dataset_numpy['components'], original_dataset['nhood'])
warnings.warn("Computing affinity labels because 'label' wasn't provided in data source.", UserWarning)
if 'mask' in original_dataset:
dataset_numpy['mask'] = get_zero_padded_array_slice(original_dataset['mask'], output_slice)
dataset_numpy['mask'] = dataset_numpy['mask'].astype(np.uint8)
else:
# assume no masking
assumed_output_mask = np.ones_like(original_dataset['components'], dtype=np.uint8)
dataset_numpy['mask'] = get_zero_padded_array_slice(assumed_output_mask, output_slice)
warnings.warn("No mask provided. Setting to 1 where outputs exist.", UserWarning)
return dataset_numpy
def test_works_with_too_positive_origin_1d(self):
X = np.ones(shape=(5,))
slice_shape = (3,)
origin = (3,)
result = get_zero_padded_slice_from_array_by_offset(X, origin, shape=slice_shape)
expected_result = np.array([1, 1, 0], dtype=X.dtype)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype
slices = [slice(o, o + s, 1) for o, s in zip(origin, slice_shape)]
result = get_zero_padded_array_slice(X, slices)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype
X = np.ones(shape=(5, 5))
slice_shape = (3, 3)
origin = (3, 3)
result = get_zero_padded_slice_from_array_by_offset(X, origin, shape=slice_shape)
expected_result = np.array([[1, 1, 0],
[1, 1, 0],
[0, 0, 0]], dtype=X.dtype)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype
slices = [slice(o, o + s, 1) for o, s in zip(origin, slice_shape)]
result = get_zero_padded_array_slice(X, slices)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype
def test_works_with_slice_bigger_than_source(self):
X = np.ones(shape=(1,))
origin = (-1,)
slice_shape = (3,)
result = get_zero_padded_slice_from_array_by_offset(X, origin, shape=slice_shape)
expected_result = np.array([0, 1, 0], dtype=X.dtype)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype
slices = [slice(o, o + s, 1) for o, s in zip(origin, slice_shape)]
result = get_zero_padded_array_slice(X, slices)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype
def test_works_with_interior_slice(self):
X = np.ones(shape=(5,))
origin = (1,)
slice_shape = (3,)
result = get_zero_padded_slice_from_array_by_offset(X, origin, shape=slice_shape)
expected_result = np.array([1, 1, 1], dtype=X.dtype)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype, "{}, {}".format(result.dtype, expected_result.dtype)
slices = [slice(o, o + s, 1) for o, s in zip(origin, slice_shape)]
result = get_zero_padded_array_slice(X, slices)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype
def test_works_with_hdf5_file(self):
with h5py.File(self.test_h5_file_name, 'r') as h5_file:
X = h5_file['test dataset']
origin = (-1, -1)
slice_shape = (3, 3)
result = get_zero_padded_slice_from_array_by_offset(X, origin, slice_shape)
expected_result = np.array([[0, 0, 0],
[0, 1, 0],
[0, 0, 0]], dtype=X.dtype)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype
slices = [slice(o, o + s, 1) for o, s in zip(origin, slice_shape)]
result = get_zero_padded_array_slice(X, slices)
assert np.allclose(result, expected_result)
assert result.dtype == expected_result.dtype
def __getitem__(self, item):
return get_zero_padded_array_slice(self.array_like, item)
