def assign_gt_to_ws(self, gt):
ws_nopad = morpho.juicy_center(self.watershed, self.pad_thickness)
bdrymap = morpho.pad(morpho.seg_to_bdry(ws_nopad),
[0] * self.pad_thickness)
gt_bdrymap_nopad = morpho.seg_to_bdry(gt)
gt_bdrymap = morpho.pad(gt_bdrymap_nopad, [0] * self.pad_thickness)
k = distance_transform_cdt(1 - bdrymap, return_indices=True)
ind = nonzero(gt_bdrymap.ravel())[0]
closest_sub = numpy.concatenate([i.ravel()[:, newaxis] for i in k[1]],
axis=1)
closest_sub = closest_sub[ind, :]
closest_ind = [
dot(bdrymap.strides, i) / bdrymap.itemsize for i in closest_sub
]
M = zeros_like(bdrymap).astype(float)
M.ravel()[closest_ind] = 1.0
bdrymap.ravel()[closest_ind] = False
k = distance_transform_cdt(1 - bdrymap, return_indices=True)
ind = nonzero(gt_bdrymap.ravel())[0]
closest_sub = numpy.concatenate([i.ravel()[:, newaxis] for i in k[1]],
axis=1)
closest_sub = closest_sub[ind, :]
closest_ind = [
dot(bdrymap.strides, i) / bdrymap.itemsize for i in closest_sub
]
M.ravel()[closest_ind] = 1.0
return M
def set_watershed(self, ws=array([]), lowmem=False, connectivity=1):
try:
self.boundary_body = ws.max() + 1
except ValueError: # empty watershed given
self.boundary_body = -1
self.volume_size = ws.size
self.has_zero_boundaries = (ws == 0).any()
if self.has_zero_boundaries:
self.watershed = morpho.pad(ws, [0, self.boundary_body])
else:
self.watershed = morpho.pad(ws, self.boundary_body)
self.segmentation = self.watershed.copy()
self.watershed_r = self.watershed.ravel()
self.segmentation_r = self.segmentation.ravel() # reduce fct calls
self.pad_thickness = 2 if (self.segmentation == 0).any() else 1
if lowmem:
def neighbor_idxs(x):
return self.get_neighbor_idxs_lean(x, connectivity)
self.neighbor_idxs = neighbor_idxs
else:
self.pixel_neighbors = \
morpho.build_neighbors_array(self.watershed, connectivity)
self.neighbor_idxs = self.get_neighbor_idxs_fast
def set_probabilities(self, probs=array([]), normalize=False):
if len(probs) == 0:
self.probabilities = zeros_like(self.watershed)
self.probabilities_r = self.probabilities.ravel()
probs = probs.astype(double)
if normalize and len(probs) > 1:
probs -= probs.min() # ensure probs.min() == 0
probs /= probs.max() # ensure probs.max() == 1
sp = probs.shape
sw = tuple(array(self.watershed.shape, dtype=int)-\
2*self.pad_thickness*ones(self.watershed.ndim, dtype=int))
p_ndim = probs.ndim
w_ndim = self.watershed.ndim
padding = [inf] + (self.pad_thickness - 1) * [0]
if p_ndim == w_ndim:
self.probabilities = morpho.pad(probs, padding)
self.probabilities_r = self.probabilities.ravel()[:, newaxis]
elif p_ndim == w_ndim + 1:
if sp[1:] == sw:
sp = sp[1:] + [sp[0]]
probs = probs.transpose(sp)
axes = range(p_ndim - 1)
self.probabilities = morpho.pad(probs, padding, axes)
self.probabilities_r = self.probabilities.reshape(
(self.watershed.size, -1))
def assign_gt_to_ws(self, gt):
ws_nopad = morpho.juicy_center(self.watershed, self.pad_thickness)
bdrymap = morpho.pad(morpho.seg_to_bdry(ws_nopad),
[0]*self.pad_thickness)
gt_bdrymap_nopad = morpho.seg_to_bdry(gt)
gt_bdrymap = morpho.pad(gt_bdrymap_nopad, [0]*self.pad_thickness)
k = distance_transform_cdt(1-bdrymap, return_indices=True)
ind = nonzero(gt_bdrymap.ravel())[0]
closest_sub = numpy.concatenate(
[i.ravel()[:,newaxis] for i in k[1]],axis=1)
closest_sub = closest_sub[ind,:]
closest_ind = [dot(bdrymap.strides, i)/bdrymap.itemsize
for i in closest_sub]
M = zeros_like(bdrymap).astype(float)
M.ravel()[closest_ind]=1.0
bdrymap.ravel()[closest_ind] = False
k = distance_transform_cdt(1-bdrymap, return_indices=True)
ind = nonzero(gt_bdrymap.ravel())[0]
closest_sub = numpy.concatenate(
[i.ravel()[:,newaxis] for i in k[1]],axis=1)
closest_sub = closest_sub[ind,:]
closest_ind = [dot(bdrymap.strides, i)/bdrymap.itemsize
for i in closest_sub]
M.ravel()[closest_ind]=1.0
return M
def set_probabilities(self, probs=array([]), normalize=False):
if len(probs) == 0:
self.probabilities = zeros_like(self.watershed)
self.probabilities_r = self.probabilities.ravel()
probs = probs.astype(double)
if normalize and len(probs) > 1:
probs -= probs.min() # ensure probs.min() == 0
probs /= probs.max() # ensure probs.max() == 1
sp = probs.shape
sw = tuple(array(self.watershed.shape, dtype=int)-\
2*self.pad_thickness*ones(self.watershed.ndim, dtype=int))
p_ndim = probs.ndim
w_ndim = self.watershed.ndim
padding = [inf]+(self.pad_thickness-1)*[0]
if p_ndim == w_ndim:
self.probabilities = morpho.pad(probs, padding)
self.probabilities_r = self.probabilities.ravel()[:,newaxis]
elif p_ndim == w_ndim+1:
if sp[1:] == sw:
sp = sp[1:]+[sp[0]]
probs = probs.transpose(sp)
axes = range(p_ndim-1)
self.probabilities = morpho.pad(probs, padding, axes)
self.probabilities_r = self.probabilities.reshape(
(self.watershed.size, -1))
def build_border(self, supervoxels1, prediction1, supervoxels2,
prediction2, mask1, mask2, reset_edges):
self.init_build(supervoxels1, prediction1)
self.init_build2(supervoxels2, prediction2)
if mask1 is not None and mask2 is not None:
mask1 = morpho.pad(mask1, 0)
mask1 = mask1.astype(numpy.double)
mask2 = morpho.pad(mask2, 0)
mask2 = mask2.astype(numpy.double)
# use masks to handle 0 cases
neuroproof.init_masks(self.stack, mask1, mask2)
self.stack.build_rag_border(reset_edges)
def build_border(self, supervoxels1, prediction1, supervoxels2,
prediction2, mask1, mask2, reset_edges):
self.init_build(supervoxels1, prediction1)
self.init_build2(supervoxels2, prediction2)
if mask1 is not None and mask2 is not None:
mask1 = morpho.pad(mask1, 0)
mask1 = mask1.astype(numpy.double)
mask2 = morpho.pad(mask2, 0)
mask2 = mask2.astype(numpy.double)
# use masks to handle 0 cases
neuroproof.init_masks(self.stack, mask1, mask2)
self.stack.build_rag_border(reset_edges)
def set_orientations(self, orientation_map, channel_is_oriented):
if len(orientation_map) == 0:
self.orientation_map = zeros_like(self.watershed)
self.orientation_map_r = self.orientation_map.ravel()
so = orientation_map.shape
sw = tuple(array(self.watershed.shape, dtype=int)-\
2*self.pad_thickness*ones(self.watershed.ndim, dtype=int))
o_ndim = orientation_map.ndim
w_ndim = self.watershed.ndim
padding = [0] + (self.pad_thickness - 1) * [0]
self.orientation_map = morpho.pad(orientation_map, padding).astype(int)
self.orientation_map_r = self.orientation_map.ravel()
if channel_is_oriented is None:
nchannels = 1 if self.probabilities.ndim==self.watershed.ndim \
else self.probabilities.shape[-1]
self.channel_is_oriented = array([False] * nchannels)
self.max_probabilities_r = zeros_like(self.probabilities_r)
self.oriented_probabilities_r = zeros_like(self.probabilities_r)
self.non_oriented_probabilities_r = self.probabilities_r
else:
self.channel_is_oriented = channel_is_oriented
self.max_probabilities_r = \
self.probabilities_r[:, self.channel_is_oriented].max(axis=1)
self.oriented_probabilities_r = \
self.probabilities_r[:, self.channel_is_oriented]
self.oriented_probabilities_r = \
self.oriented_probabilities_r[
range(len(self.oriented_probabilities_r)),
self.orientation_map_r]
self.non_oriented_probabilities_r = \
self.probabilities_r[:, ~self.channel_is_oriented]
def set_orientations(self, orientation_map, channel_is_oriented):
if len(orientation_map) == 0:
self.orientation_map = zeros_like(self.watershed)
self.orientation_map_r = self.orientation_map.ravel()
so = orientation_map.shape
sw = tuple(array(self.watershed.shape, dtype=int)-\
2*self.pad_thickness*ones(self.watershed.ndim, dtype=int))
o_ndim = orientation_map.ndim
w_ndim = self.watershed.ndim
padding = [0]+(self.pad_thickness-1)*[0]
self.orientation_map = morpho.pad(orientation_map, padding).astype(int)
self.orientation_map_r = self.orientation_map.ravel()
if channel_is_oriented is None:
nchannels = 1 if self.probabilities.ndim==self.watershed.ndim \
else self.probabilities.shape[-1]
self.channel_is_oriented = array([False]*nchannels)
self.max_probabilities_r = zeros_like(self.probabilities_r)
self.oriented_probabilities_r = zeros_like(self.probabilities_r)
self.non_oriented_probabilities_r = self.probabilities_r
else:
self.channel_is_oriented = channel_is_oriented
self.max_probabilities_r = \
self.probabilities_r[:, self.channel_is_oriented].max(axis=1)
self.oriented_probabilities_r = \
self.probabilities_r[:, self.channel_is_oriented]
self.oriented_probabilities_r = \
self.oriented_probabilities_r[
range(len(self.oriented_probabilities_r)),
self.orientation_map_r]
self.non_oriented_probabilities_r = \
self.probabilities_r[:, ~self.channel_is_oriented]
def init_build2(self, watershed, probabilities):
watershed = morpho.pad(watershed, 0)
watershed = watershed.astype(numpy.double)
neuroproof.reinit_stack2(self.stack, watershed)
probabilities = probabilities.astype(numpy.double)
num_channels = 1
if self.single_channel:
probabilities = morpho.pad(probabilities, 0)
neuroproof.add_prediction_channel2(self.stack, probabilities)
else:
num_channels = probabilities.shape[probabilities.ndim - 1]
for channel in range(0, num_channels):
curr_prob = morpho.pad(probabilities[..., channel], 0)
neuroproof.add_prediction_channel2(self.stack, curr_prob)
def init_build2(self, watershed, probabilities):
watershed = morpho.pad(watershed, 0)
watershed = watershed.astype(numpy.double)
neuroproof.reinit_stack2(self.stack, watershed)
probabilities = probabilities.astype(numpy.double)
num_channels = 1
if self.single_channel:
probabilities = morpho.pad(probabilities, 0)
neuroproof.add_prediction_channel2(self.stack, probabilities)
else:
num_channels = probabilities.shape[probabilities.ndim-1]
for channel in range(0,num_channels):
curr_prob = morpho.pad(probabilities[...,channel], 0)
neuroproof.add_prediction_channel2(self.stack, curr_prob)
def remove_merged_boundaries(ar):
import morpho
arp = morpho.pad(ar, [0,ar.max()+1])
arpr = arp.ravel()
zero_idxs = (arpr == 0).nonzero()[0]
ns = arpr[morpho.get_neighbor_idxs(arp, zero_idxs)]
ns_compl = ns.copy()
ns_compl[ns==0] = ns.max()+1
merged_boundaries = (ns.max(axis=1) == ns_compl.min(axis=1)).nonzero()[0]
arpr[zero_idxs[merged_boundaries]] = ns.max(axis=1)[merged_boundaries]
return morpho.juicy_center(arp, 2)
def remove_merged_boundaries(ar):
import morpho
arp = morpho.pad(ar, [0, ar.max() + 1])
arpr = arp.ravel()
zero_idxs = (arpr == 0).nonzero()[0]
ns = arpr[morpho.get_neighbor_idxs(arp, zero_idxs)]
ns_compl = ns.copy()
ns_compl[ns == 0] = ns.max() + 1
merged_boundaries = (ns.max(axis=1) == ns_compl.min(axis=1)).nonzero()[0]
arpr[zero_idxs[merged_boundaries]] = ns.max(axis=1)[merged_boundaries]
return morpho.juicy_center(arp, 2)
def set_watershed(self, ws=array([]), lowmem=False, connectivity=1):
try:
self.boundary_body = ws.max()+1
except ValueError: # empty watershed given
self.boundary_body = -1
self.volume_size = ws.size
self.has_zero_boundaries = (ws==0).any()
if self.has_zero_boundaries:
self.watershed = morpho.pad(ws, [0, self.boundary_body])
else:
self.watershed = morpho.pad(ws, self.boundary_body)
self.segmentation = self.watershed.copy()
self.watershed_r = self.watershed.ravel()
self.segmentation_r = self.segmentation.ravel() # reduce fct calls
self.pad_thickness = 2 if (self.segmentation==0).any() else 1
if lowmem:
def neighbor_idxs(x):
return self.get_neighbor_idxs_lean(x, connectivity)
self.neighbor_idxs = neighbor_idxs
else:
self.pixel_neighbors = \
morpho.build_neighbors_array(self.watershed, connectivity)
self.neighbor_idxs = self.get_neighbor_idxs_fast
def set_ground_truth(self, gt=None):
if gt is not None:
gtm = gt.max() + 1
gt_ignore = [0, gtm] if (gt == 0).any() else [gtm]
seg_ignore = [0, self.boundary_body] if \
(self.segmentation==0).any() else [self.boundary_body]
self.gt = morpho.pad(gt, gt_ignore)
self.rig = contingency_table(self.segmentation, self.gt)
self.rig[:, gt_ignore] = 0
self.rig[seg_ignore, :] = 0
else:
self.gt = None
# null pattern to transparently allow merging of nodes.
# Bonus feature: counts how many sp's went into a single node.
try:
self.rig = ones(self.watershed.max() + 1)
except ValueError:
self.rig = ones(self.number_of_nodes() + 1)
def set_ground_truth(self, gt=None):
if gt is not None:
gtm = gt.max()+1
gt_ignore = [0, gtm] if (gt==0).any() else [gtm]
seg_ignore = [0, self.boundary_body] if \
(self.segmentation==0).any() else [self.boundary_body]
self.gt = morpho.pad(gt, gt_ignore)
self.rig = contingency_table(self.segmentation, self.gt)
self.rig[:, gt_ignore] = 0
self.rig[seg_ignore, :] = 0
else:
self.gt = None
# null pattern to transparently allow merging of nodes.
# Bonus feature: counts how many sp's went into a single node.
try:
self.rig = ones(self.watershed.max()+1)
except ValueError:
self.rig = ones(self.number_of_nodes()+1)
