def construct_from_parameters_add_tree():
parameters = Skeleton.define_parameters('2017-01-12_FD0156-2',
(11.24, 11.24, 32))
skel = Skeleton(parameters=parameters)
nodes = skel.define_nodes([40000, 40100, 40200], [45000, 45100, 45200],
[1000, 1100, 1200], [1, 2, 3])
edges = [(1, 3), (2, 3)]
skel.add_tree(nodes, edges)
skel.write_nml('testdata/PA_gen.nml')
def add_nodes_as_trees():
# Test merging skeleton with multiple trees from nodes
skel = Skeleton('testdata/02_ref.nml')
nodes = skel.define_nodes([40000, 40100, 40200], [45000, 45100, 45200],
[1000, 1100, 1200])
skel.add_nodes_as_trees(nodes)
skel.write_nml('testdata/01_NT_merged_gen.nml')
def add_tree_from_skel():
# Test merging skeleton with multiple trees with a single tree from another skeleton
skel = Skeleton('testdata/01_ref.nml')
skel.add_tree_from_skel(Skeleton('testdata/02_ref.nml'),
tree_idx=1,
group_id=10)
# Test for unique node and tree ids in merged nml
_, c = np.unique(np.concatenate([nodes.id.values for nodes in skel.nodes]),
return_counts=True)
assert all(c < 2)
_, c = np.unique(skel.tree_ids, return_counts=True)
assert all(c < 2)
# Test writing
skel.write_nml('testdata/01_02_single_merged_gen.nml')
cache_HDD_root = os.path.join(path_in, '.cache/')
path_datasources = os.path.join(path_in, 'datasources.json')
path_nml_in = os.path.join(path_in, 'bbox_annotated.nml')
input_shape = (140, 140, 1)
target_shape = (1, 1, 1)
stride = (35, 35, 1)
datasources = WkwData.datasources_from_json(path_datasources)
dataset = WkwData(input_shape=input_shape,
target_shape=target_shape,
data_sources=datasources,
stride=stride,
cache_HDD=False,
cache_RAM=True)
skel = Skeleton(path_nml_in)
pred_df = pd.DataFrame(columns=[
'tree_idx', 'tree_id', 'x', 'y', 'z', 'xi', 'yi', 'class', 'explicit',
'cluster_id', 'prob'
])
group_ids = np.array(skel.group_ids)
input_path = datasources[0].input_path
input_bbox = datasources[0].input_bbox
structure = np.ones((3, 3), dtype=np.int)
cluster_id = 0
for plane_group in skel.groups:
plane_group_id = plane_group.id
plane_group_class = bool(int(plane_group.name[-1]))
plane_tree_inds = np.where(group_ids == plane_group_id)[0]
plane_matrix = np.zeros((5, 5), dtype=np.bool)
def get_distances_to_node():
skel = Skeleton('testdata/01_ref.nml')
positions = skel.nodes[5].position
skel.get_distances_to_node(positions=positions, node_id=35370)
skel.get_distances_to_node(positions=positions, tree_idx=0, node_idx=5)
def get_distance_to_nodes():
skel = Skeleton('testdata/01_ref.nml')
skel.get_distance_to_nodes(position=(40000, 38000, 1000), tree_idx=6)
from genEM3.util.image import bboxesFromArray
from genEM3.data.skeleton import get_volume_df
# Get the names of three test skeletons
path_in_stub = os.path.join(get_runs_dir(),
'inference/ae_classify_11_parallel')
test_dirs = ['test_center_filt', 'test_bottom_filt', 'test_top_filt']
skel_dirs = [
os.path.join(path_in_stub, d, 'bbox_annotated.nml') for d in test_dirs
]
# check that all of the files exist
assert all([os.path.exists(skel_dir) for skel_dir in skel_dirs])
# Create skeleton objects
start = time.time()
skeletons = [Skeleton(skel_dir) for skel_dir in skel_dirs]
print(f'Time to read skeleton: {time.time() - start}')
# Read the coordinates and target class of all three skeletons into the volume data frame
volume_df = get_volume_df(skeletons=skeletons)
# Get the ingredients for making the datasources
bboxes = bboxesFromArray(volume_df[['x', 'y', 'z']].values)
input_dir = '/tmpscratch/webknossos/Connectomics_Department/2018-11-13_scMS109_1to7199_v01_l4_06_24_fixed_mag8_artifact_pred/color/1'
target_class = volume_df['class'].values.astype(np.float)
target_binary = 1
target_dir = input_dir
input_mean = 148.0
input_std = 36.0
# Create a list of data sources
source_list = []
for i, cur_bbox in enumerate(bboxes):
cur_target = target_class[i]
def add_trees_from_skel():
# Test merging skeletons both having both root and (nested) group trees
skel = Skeleton('testdata/01_ref.nml')
assert all([
Skeleton._num_conn_comp(Skeleton._get_graph(nodes, edges)) == 1
for nodes, edges in zip(skel.nodes, skel.edges)
])
skel.add_trees_from_skel(Skeleton('testdata/02_ref.nml'))
# Test for unique node and tree ids in merged nml
_, c = np.unique(np.concatenate([nodes.id.values for nodes in skel.nodes]),
return_counts=True)
assert all(c < 2)
_, c = np.unique(skel.tree_ids, return_counts=True)
assert all(c < 2)
# Test writing
assert all([
Skeleton._num_conn_comp(Skeleton._get_graph(nodes, edges)) == 1
for nodes, edges in zip(skel.nodes, skel.edges)
])
skel.write_nml('testdata/01_02_merged_gen.nml')
# Test reading
skel_gen = Skeleton('testdata/01_02_merged_gen.nml')
skel_ref = Skeleton('testdata/01_02_merged_ref.nml')
# Test for equal numbers of nodes and edges in both generated and reference merge nml
assert set([len(nodes) for nodes in skel_gen.nodes
]) == set([len(nodes) for nodes in skel_ref.nodes])
assert set([len(edges) for edges in skel_gen.edges
]) == set([len(edges) for edges in skel_ref.edges])
# Test merging skeleton having both root and group trees with one having only root trees
skel = Skeleton('testdata/02_ref.nml')
skel.add_trees_from_skel(Skeleton('testdata/03_ref.nml'))
skel.write_nml('testdata/02_03_merged_gen.nml')
# Test merging skeleton having only root trees with one having both root and group trees
skel = Skeleton('testdata/03_ref.nml')
skel.add_trees_from_skel(Skeleton('testdata/02_ref.nml'))
skel.write_nml('testdata/03_02_merged_gen.nml')
data_sources=datasources,
stride=stride,
cache_HDD=False,
cache_RAM=False)
bbox_val_dims_vx = np.array(bbox_val_dims_um) * 1000 / np.array(scale)
n_fits = np.ceil((bbox_val_dims_vx - np.array(input_shape)) / np.array(stride)).astype(int)
meshes = dataset.data_meshes[0]['input']
meshes_shape = np.array(meshes['x'].shape)
meshes_center = np.floor(meshes_shape / 2)
meshes_min = np.floor(meshes_center - n_fits/2).astype(int)
meshes_max = np.floor(meshes_center + n_fits/2).astype(int)
meshes_val = {key: meshes[key][meshes_min[0]:meshes_max[0], meshes_min[1]:meshes_max[1], meshes_min[2]:meshes_max[2]] for key in meshes}
skel = Skeleton(os.path.join(get_runs_dir(), 'inference/ae_classify_11_parallel/empty.nml'))
min_id = 1
for idx in range(np.prod(n_fits)):
print('adding trees {}/{}'.format(idx, np.prod(n_fits)))
xi, yi, zi = np.unravel_index(idx, shape=n_fits)
cx = meshes_val['x'][xi, yi, zi]
cy = meshes_val['y'][xi, yi, zi]
cz = meshes_val['z'][xi, yi, zi]
positions = np.array([
[cx, cy, cz],
[cx - input_shape[0]/2, cy - input_shape[1]/2, cz],
[cx - input_shape[0]/2, cy + input_shape[1]/2, cz],
[cx + input_shape[0]/2, cy - input_shape[1]/2, cz],
[cx + input_shape[0]/2, cy + input_shape[1]/2, cz],
[cx + input_shape[0]/2, cy + input_shape[1]/2 - 1, cz]
def construct_from_parameters_empty():
parameters = Skeleton.define_parameters('2017-01-12_FD0156-2',
(11.24, 11.24, 32))
skel = Skeleton(parameters=parameters)
skel.write_nml('testdata/PE_gen.nml')
def construct_from_nml():
skel = Skeleton('testdata/02_ref.nml')
skel.write_nml('testdata/02_gen.nml')
def plot():
skel = Skeleton('testdata/01_02_merged_ref.nml')
skel.plot(unit='um', view=None, colors='Dark2')