def test_add_factors_generic(self):
def mygen():
yield 0
yield 1
gm = opengm.gm([2, 4])
f = opengm.PottsFunction([2, 4], 0.0, 1.0)
fid = gm.addFunction(f)
vis_list = [
[0, 1],
(0, 1),
(x for x in xrange(2)),
mygen(),
opengm.IndexVector(x for x in xrange(0, 2)),
numpy.arange(0, 2, dtype=numpy.uint64),
]
for i, vis in enumerate(vis_list):
fIndex = gm.addFactor(fid, vis)
assert gm.numberOfFactors == i + 1
assert fIndex == i
assert gm[fIndex].numberOfVariables == 2
assert gm[fIndex].shape[0] == 2
assert gm[fIndex].shape[1] == 4
assert gm[fIndex].variableIndices[0] == 0
assert gm[fIndex].variableIndices[1] == 1
def create_graph(gm_db):
print("Creating Graph ...")
all_name_pairs = gm_db['pairwises'].keys()
all_names = set()
for n1, n2 in all_name_pairs:
all_names.add(n1)
all_names.add(n2)
node2id = {}
id2node = {}
num_labels = gm_db['pairwises'][all_name_pairs[0]].shape[0]
for i, n in enumerate(all_names):
node2id[n] = i
id2node[i] = n
num_nodes = len(all_names)
print("Num nodes: {}".format(num_nodes))
gm = opengm.gm([num_labels] * num_nodes, operator='adder')
if len(gm_db['unaries']) > 0:
for i in range(num_nodes):
gm.addFactor(gm.addFunction(gm_db['unaries'][id2node[i]]), [i])
for n1, n2 in all_name_pairs:
id1 = node2id[n1]
id2 = node2id[n2]
t = [id1, id2]
p = gm_db['pairwises'][(n1, n2)]
if id2 < id1:
t = [id2, id1]
p = p.T
if id1 == id2:
continue
gm.addFactor(gm.addFunction(p), t)
print(" - Done")
return gm, node2id, id2node
def onClickedMulticut(self):
p1 = self.probs.copy()
p1 = numpy.clip(p1, 0.005, 1.0-0.005)
p0 = 1.0 - p1
weights = numpy.log(p0/p1)
nVar = self.rag.maxNodeId + 1
nos = numpy.ones(nVar)*nVar
gm = opengm.gm(nos)
uv = self.rag.uvIds()
uv = numpy.sort(uv,axis=1)
pf = opengm.pottsFunctions([nVar,nVar], numpy.array([0]),weights)
fid = gm.addFunctions(pf)
gm.addFactors(fid,uv)
pparam = opengm.InfParam(seedFraction=0.05)
parameter = opengm.InfParam(generator='randomizedWatershed',proposalParam=pparam,numStopIt=10,numIt=3000)
inf = opengm.inference.IntersectionBased(gm, parameter=parameter)
inf = opengm.inference.Multicut(gm)
inf.infer(inf.verboseVisitor())
arg = inf.arg()
self.eArg = arg[uv[:,0]]!=arg[uv[:,1]]
self.ctrlWidget.modeSelectorComboBox.setCurrentIndex(6)
def test_add_multiple_functions_order2a(self):
nVar = 4
nLabels = 2
gm = opengm.gm([nLabels] * nVar)
# add functionS
fShape =[2,2,2]
f = opengm.randomFunction(fShape)
vis=numpy.ones([4,2])
vis[0,0]=0
vis[0,1]=1
vis[1,0]=1
vis[1,1]=2
fid = gm.addFunction(f)
gm.addFactor(fid,[0,1,2])
assert gm[0][0,0,0]==f[0,0,0]
assert gm[0][1,0,0]==f[1,0,0]
assert gm[0][0,1,0]==f[0,1,0]
assert gm[0][1,1,0]==f[1,1,0]
def generate_grid(dimx, dimy, labels, beta1, beta2, operator="adder"):
nos = numpy.ones(dimx * dimy, dtype=numpy.uint64) * labels
gm = opengm.gm(nos, operator, 0)
for vi in range(dimx * dimy):
f1 = numpy.random.random((labels,)).astype(numpy.float64) * 0.6 + 0.2
assert len(f1.shape) == 1
assert f1.shape[0] == labels
fid1 = gm.addFunction(f1)
gm.addFactor(fid1, (vi,))
f2 = numpy.ones([labels, labels], dtype=numpy.float64)
for l in range(labels):
f2[l, l] = beta1
fid2 = gm.addFunction(f2)
for y in range(dimy):
for x in range(dimx):
if x + 1 < dimx:
vis = [x + y * dimx, x + 1 + y * dimx]
assert vis.sort is not None
vis.sort
gm.addFactor(fid2, vis)
if y + 1 < dimy:
vis = [x + y * dimx, x + (y + 1) * dimx]
vis.sort()
gm.addFactor(fid2, vis)
return gm
def test_add_multiple_functions_order1(self):
nVar = 4
nLabels = 2
gm = opengm.gm([nLabels] * nVar)
# add functionS
fShape =[4,2]
f = opengm.randomFunction(fShape)
vis=numpy.ones([4,1])
vis[0,0]=0
vis[1,0]=1
vis[2,0]=2
vis[3,0]=3
fids = gm.addFunctions(f)
gm.addFactors(fids,vis)
assert gm[1][(0,)]==f[1,0]
assert gm[1][(1,)]==f[1,1]
for x in xrange(4):
assert gm[x][(0,)]==f[x,0]
assert gm[x][(1,)]==f[x,1]
def instantiate_sentence(sentence_data):
# Get the number of tokens:
numVar = len(sentence_data)
# print 'numVar', numVar
# The domain of the variables is the len of th y_list
# Use the multiplier, the adder doesn't seem to work
gm = opengm.gm([len(y_list)] * numVar, "multiplier")
# gm=opengm.gm([len(y_list)]*numVar,'adder')
# print gm.numberOfVariables
# exit()
# Create the 1st factors
gm = create_first_order(gm, sentence_data)
# print gm.factors()
# for fac in gm.factors():
# print fac
# Create the 2nd order factors (True for shared factors)
gm = create_second_order(gm, sentence_data, True)
# print gm.factors()
# for fac in gm.factors():
# print fac
# exit()
return gm
def segment(self,weights,warmStart=None,verbose=False): #try : # self.cgc.changeWeights(weights) #except : nVar = self.cgp.numCells(2) nFac = self.cgp.numCells(1) cell1Bounds = self.cgp.cell1BoundsArray()-1 self.gm2 = opengm.gm(numpy.ones(nVar,dtype=opengm.label_type)*nVar) assert self.gm2.numberOfVariables == nVar # init with zero potts functions #print weights pf = opengm.pottsFunctions([nVar,nVar],numpy.zeros(nFac),weights ) fids = self.gm2.addFunctions(pf) # add factors self.gm2.addFactors(fids,cell1Bounds) self.cgc2 = opengm.inference.Cgc(gm=self.gm2,parameter=opengm.InfParam(planar=True)) if verbose : self.cgc2.infer(self.cgc.verboseVisitor()) else : self.cgc2.infer() self.labels[:]=self.cgc2.arg()
def __init__(self,cgp,beta=0.5):
self.cgp=cgp
self.probability = numpy.ones(cgp.numCells(1),dtype=numpy.float64)
self.weights = numpy.ones(cgp.numCells(1),dtype=numpy.float64)
self.mean = numpy.ones(cgp.numCells(1),dtype=numpy.float64)
self.std = numpy.ones(cgp.numCells(1),dtype=numpy.float64)
# generate graphical model
self.cgc = None
self.beta=beta
boundArray = self.cgp.cell1BoundsArray()-1
nVar = cgp.numCells(2)
nFac = cgp.numCells(1)
space = numpy.ones(nVar,dtype=opengm.label_type)*nVar
self.gm = opengm.gm(space)
wZero = numpy.zeros(nFac,dtype=opengm.value_type)
pf=opengm.pottsFunctions([nVar,nVar],wZero,wZero)
fids = self.gm.addFunctions(pf)
self.gm.addFactors(fids,boundArray)
self.cgc = opengm.inference.Cgc(gm=self.gm,parameter=opengm.InfParam(planar=True))
def predict():
global features,rag, sp,arg
trainingInstances = numpy.array(userLabels.keys(),dtype='uint64')
labels = numpy.array([userLabels[k] for k in trainingInstances],dtype='uint32')[:,None]
trainingInstances = numpy.clip(trainingInstances, 0, rag.maxEdgeId-2)
assert trainingInstances.max() <= rag.maxEdgeId
if len(labels)>10 and labels.min()==0 and labels.max()==1:
feat = features[trainingInstances,:]
rf = vigra.learning.RandomForest(treeCount=255)
oob = rf.learnRF(feat, labels)
print "oob", oob
probs = rf.predictProbabilities(features)[:,1]
p1 = probs.copy()
p1 = numpy.clip(p1, 0.005, 1.0-0.005)
p0 = 1.0 - p1
weights = numpy.log(p0/p1)
nVar = rag.maxNodeId + 1
nos = numpy.ones(nVar)*nVar
gm = opengm.gm(nos)
uv = rag.uvIds()
if weights.shape[0] < uv.shape[1]:
diff = uv.shape[1] - weights.shape[0]
val = numpy.zeros(diff)
weights = numpy.concatenate([weights,val])
uv = numpy.sort(uv,axis=1)
pf = opengm.pottsFunctions([nVar,nVar], numpy.array([0]),weights)
fid = gm.addFunctions(pf)
gm.addFactors(fid,uv)
param = opengm.InfParam(planar=False)
inf = opengm.inference.Cgc(gm,parameter=param)
if arg is not None:
inf.setStartingPoint(arg)
visitor = inf.timingVisitor(timeLimit=60.0)
inf.infer(visitor)
arg = inf.arg()
eArg = arg[uv[:,0]]!=arg[uv[:,1]]
return arg
else:
return None
def infer(self, pairwise_scores, unary_scores, k, mrf_type, filenames, target_class, bcd_scores):
'''
Returns:
- argmin: a list of size k where each element is the label selected for the
corresponding node.
- t: runing time for inference
'''
assert(bcd_scores is not None)
import opengm
num_internal_nodes = self.num_nodes
num_external_nodes = bcd_scores.shape[0]
num_labels = int(np.sqrt(pairwise_scores.shape[1]))
indices = self.get_indices()
gm = opengm.gm([num_labels]*num_internal_nodes + [1]*num_external_nodes, operator='adder')
if unary_scores is not None:
assert unary_scores.shape[0] == k
assert unary_scores.shape[1] == num_labels
for i in range(num_internal_nodes):
gm.addFactor(gm.addFunction(-unary_scores[i]), [i])
for i,e in enumerate(indices):
gm.addFactor(gm.addFunction(-pairwise_scores[i].reshape(
(num_labels, num_labels))), [e[0],e[1]])
for i,j in itertools.product(range(num_external_nodes), range(num_internal_nodes)):
gm.addFactor(gm.addFunction(-bcd_scores[i,j][...,None]),
[j, i+num_internal_nodes])
if mrf_type.endswith('astar'):
inf = opengm.inference.AStar(gm, accumulator='minimizer')
elif mrf_type.endswith('trws'):
inf = opengm.inference.TrwsExternal(gm, accumulator='minimizer')
elif mrf_type.endswith('trbp'):
inf = opengm.inference.TreeReweightedBp(gm, accumulator='minimizer')
elif mrf_type.endswith('map'):
inf = opengm.inference.BeliefPropagation(gm, accumulator='minimizer')
else:
raise ValueError('Inference for mrf type {} is not implemented'.format(mrf_type))
#br_inf = opengm.inference.Bruteforce(gm, accumulator='minimizer')
#br_inf.infer()
t0 = time.time()
inf.infer()
argmin = inf.arg()
t1 = time.time() - t0
energy = inf.value()
self._debug = dict(gm=gm, pairwise_scores=pairwise_scores,
unary_scores=unary_scores,
k=k, mrf_type=mrf_type,
filenames=filenames, target_class=target_class,
bcd_scores=bcd_scores,
argmin=argmin,
min_energy=energy)
return argmin[:k], t1, energy
def intra_encounter_matching():
import numpy as np
from scipy.sparse import coo_matrix, csgraph
qreq_, cm_list = testdata_workflow()
# qaids = [cm.qaid for cm in cm_list]
# top_aids = [cm.get_top_aids(5) for cm in cm_list]
aid_pairs = np.array([(cm.qaid, daid)
for cm in cm_list for daid in cm.get_top_aids(5)])
top_scores = ut.flatten([cm.get_top_scores(5) for cm in cm_list])
N = aid_pairs.max() + 1
mat = coo_matrix((top_scores, aid_pairs.T), shape=(N, N))
csgraph.connected_components(mat)
tree = csgraph.minimum_spanning_tree(mat) # NOQA
import plottool as pt
dense = mat.todense()
pt.imshow(dense / dense.max() * 255)
pt.show_if_requested()
# baseline jobid
import opengm
# https://github.com/opengm/opengm/blob/master/src/interfaces/python/examples/tutorial/OpenGM%20tutorial.ipynb
numVar = 10
unaries = np.ones([numVar, 3], dtype=opengm.value_type)
gm = opengm.gm(np.ones(numVar, dtype=opengm.label_type) * 3)
unary_fids = gm.addFunctions(unaries)
gm.addFactors(unary_fids, np.arange(numVar))
infParam = opengm.InfParam(
workflow=ut.ensure_ascii('(IC)(TTC-I,CC-I)'),
)
inf = opengm.inference.Multicut(gm, parameter=infParam)
visitor = inf.verboseVisitor(printNth=1, multiline=False)
inf.infer(visitor)
arg = inf.arg()
# gridVariableIndices = opengm.secondOrderGridVis(img.shape[0], img.shape[1])
# fid = gm.addFunction(regularizer)
# gm.addFactors(fid, gridVariableIndices)
# regularizer = opengm.pottsFunction([3, 3], 0.0, beta)
# gridVariableIndices = opengm.secondOrderGridVis(img.shape[0], img.shape[1])
# fid = gm.addFunction(regularizer)
# gm.addFactors(fid, gridVariableIndices)
unaries = np.random.rand(10, 10, 2)
potts = opengm.PottsFunction([2, 2], 0.0, 0.4)
gm = opengm.grid2d2Order(unaries=unaries, regularizer=potts)
inf = opengm.inference.GraphCut(gm)
inf.infer()
arg = inf.arg() # NOQA
"""
def test_logistic_as_gm():
# Define the number or features:
n_features = 5000
# Initialize the FeatureFunction object:
FF = FeatFunctions(n_features=5000)
# Create a weights vector, assinging a weight to each feature:
global theta
theta = 0.5 * randn(n_features)
# Get a sentence:
s0 = 'esto esta chido'
# Define the feature set for this factor:
# feat_list_names_factor0 = ['url','all_caps','ngrams']
feat_list_names_factor0 = ['ngrams']
# Define the functions for the factors:
# phi0 = funCreator('phi0','I',feat_list_names_factor0,s0,FF)
phi0 = funCreator('phi0',['I'],feat_list_names_factor0,s0,FF)
# Initalize a graphical model:
# For this example, one single variable, logistic regression
numVars = 1
cardinality = [3] # Binary classifier
# Create the gm:
gm=opengm.gm(cardinality*numVars,'multiplier')
# Transform the function to opengm:
# The second parameter is a list of the cardinalities of the variables in the scope of this function
py_func_phi0 = opengm.PythonFunction(phi0,[3])
# Add the function the pgm
gm_func_phi0 = gm.addFunction(py_func_phi0)
# # Add the opengm function to the gm model
# The second parameter is the [set] of variables in the scope of this factor
gm.addFactor(gm_func_phi0,0)
# Run inference to get the variable marginals:
bp = opengm.inference.BeliefPropagation(gm, 'integrator')
bp.infer()
# Get a list of all variables in the sentence:
var_idx = [x for x in xrange(gm.numberOfVariables)]
marg = bp.marginals(var_idx)
print 'marg:\n',marg
def cut_step(G, nodes, edges, n_annots, n_names, lookup_annot_idx, edge_probs, pass_values, fail_values):
# Create nodes in the graphical model. In this case there are <num_vars>
# nodes and each node can be assigned to one of <num_vars> possible labels
space = np.full((n_annots,), fill_value=n_names, dtype=opengm.index_type)
gm = opengm.gm(space, operator='adder')
# Use one potts function for each edge
gm = build_factor_graph(G, nodes, edges , n_annots, n_names,
lookup_annot_idx, use_unaries=False,
edge_probs=edge_probs, operator='adder')
with ut.Indenter('[CUTS]'):
ut.cprint('Brute Force Labels: (energy minimization)', 'blue')
infr = opengm.inference.Bruteforce(gm, accumulator='minimizer')
infr.infer()
labels = rectify_labels(G, infr.arg())
print(pd.DataFrame(labels, columns=['nid'], index=pd.Series(nodes)).T)
print('value = %r' % (infr.value(),))
mc_params = opengm.InfParam(maximalNumberOfConstraintsPerRound=1000000,
initializeWith3Cycles=True,
edgeRoundingValue=1e-08, timeOut=36000000.0,
cutUp=1e+75, reductionMode=3, numThreads=0,
# allowCutsWithin=?
# workflow=workflow
verbose=False, verboseCPLEX=False)
infr = opengm.inference.Multicut(gm, parameter=mc_params,
accumulator='minimizer')
infr.infer()
labels = infr.arg()
labels = rectify_labels(G, infr.arg())
ut.cprint('Multicut Labels: (energy minimization)', 'blue')
print(pd.DataFrame(labels, columns=['nid'], index=pd.Series(nodes)).T)
print('value = %r' % (infr.value(),))
if pass_values is not None:
gotany = False
for pval in pass_values:
if all(labels == pval):
gotany = True
break
if not gotany:
ut.cprint('INCORRECT DID NOT GET PASS VALUES', 'red')
print('pass_values = %r' % (pass_values,))
if fail_values is not None:
for fail in fail_values:
if all(labels == fail):
ut.cprint('INCORRECT', 'red')
def test_add_multiple_functions_with_map(self):
gm = opengm.gm([2] * 10)
def add_a_function(w):
return gm.addFunction(opengm.differenceFunction(shape=[2, 2], weight=w))
weights = [0.2, 0.3, 0.4]
fidList = map(add_a_function, weights)
assert isinstance(fidList, list)
assert len(fidList) == len(weights)
gm.addFactors(fidList, [[0, 1], [1, 2], [3, 4]])
def test_add_multiple_functions(self):
nVar = 10
nLabels = 2
for nFunctions in [1, 10]:
for order in [1, 2, 3, 4]:
gm = opengm.gm([nLabels] * nVar)
# add functionS
fShape = [nFunctions] + [nLabels] * order
f = numpy.ones(fShape, dtype=opengm.value_type).reshape(-1)
f[:] = numpy.random.rand(f.size)[:]
f = f.reshape(fShape)
fids = gm.addFunctions(f)
# assertions
assert len(fids) == nFunctions
def intra_encounter_matching():
import numpy as np
from scipy.sparse import coo_matrix, csgraph
qreq_, cm_list = testdata_workflow()
# qaids = [cm.qaid for cm in cm_list]
# top_aids = [cm.get_top_aids(5) for cm in cm_list]
aid_pairs = np.array([(cm.qaid, daid) for cm in cm_list
for daid in cm.get_top_aids(5)])
top_scores = ut.flatten([cm.get_top_scores(5) for cm in cm_list])
N = aid_pairs.max() + 1
mat = coo_matrix((top_scores, aid_pairs.T), shape=(N, N))
csgraph.connected_components(mat)
tree = csgraph.minimum_spanning_tree(mat) # NOQA
import plottool as pt
dense = mat.todense()
pt.imshow(dense / dense.max() * 255)
pt.show_if_requested()
# baseline jobid
import opengm
# https://github.com/opengm/opengm/blob/master/src/interfaces/python/examples/tutorial/OpenGM%20tutorial.ipynb
numVar = 10
unaries = np.ones([numVar, 3], dtype=opengm.value_type)
gm = opengm.gm(np.ones(numVar, dtype=opengm.label_type) * 3)
unary_fids = gm.addFunctions(unaries)
gm.addFactors(unary_fids, np.arange(numVar))
infParam = opengm.InfParam(workflow=ut.ensure_ascii('(IC)(TTC-I,CC-I)'), )
inf = opengm.inference.Multicut(gm, parameter=infParam)
visitor = inf.verboseVisitor(printNth=1, multiline=False)
inf.infer(visitor)
arg = inf.arg()
# gridVariableIndices = opengm.secondOrderGridVis(img.shape[0], img.shape[1])
# fid = gm.addFunction(regularizer)
# gm.addFactors(fid, gridVariableIndices)
# regularizer = opengm.pottsFunction([3, 3], 0.0, beta)
# gridVariableIndices = opengm.secondOrderGridVis(img.shape[0], img.shape[1])
# fid = gm.addFunction(regularizer)
# gm.addFactors(fid, gridVariableIndices)
unaries = np.random.rand(10, 10, 2)
potts = opengm.PottsFunction([2, 2], 0.0, 0.4)
gm = opengm.grid2d2Order(unaries=unaries, regularizer=potts)
inf = opengm.inference.GraphCut(gm)
inf.infer()
arg = inf.arg() # NOQA
"""
def test_constructor_generic(self):
def mygen():
yield 2
yield 3
yield 4
nos_list = [
numpy.arange(2, 5, dtype=numpy.uint64),
[2, 3, 4],
(2, 3, 4),
(x for x in xrange(2, 5)),
mygen(),
opengm.IndexVector(x for x in xrange(2, 5))
]
for i, nos in enumerate(nos_list):
if(type(nos) != type(mygen())):
pass
# assert(len(nos)==3)
gm = opengm.gm(nos, operator='adder')
assert(gm.numberOfVariables == 3)
assert(gm.numberOfLabels(0) == 2)
assert(gm.numberOfLabels(1) == 3)
assert(gm.numberOfLabels(2) == 4)
assert(gm.space().numberOfVariables == 3)
assert(gm.space()[0] == 2)
assert(gm.space()[1] == 3)
assert(gm.space()[2] == 4)
nos_list = [
numpy.arange(2, 5, dtype=numpy.uint64),
[2, 3, 4],
(2, 3, 4),
(x for x in xrange(2, 5)),
mygen(),
opengm.IndexVector(x for x in xrange(2, 5))
]
for i, nos in enumerate(nos_list):
if(type(nos) != type(mygen())):
pass # assert(len(nos)==3)
gm = opengm.adder.GraphicalModel()
gm.assign(nos)
assert(gm.numberOfVariables == 3)
assert(gm.numberOfLabels(0) == 2)
assert(gm.numberOfLabels(1) == 3)
assert(gm.numberOfLabels(2) == 4)
assert(gm.space().numberOfVariables == 3)
assert(gm.space()[0] == 2)
assert(gm.space()[1] == 3)
assert(gm.space()[2] == 4)
def test_add_multiple_functions_with_map(self):
gm = opengm.gm([2] * 10)
def add_a_function(w):
return gm.addFunction(opengm.differenceFunction(shape=[2, 2],
weight=w))
weights = [0.2, 0.3, 0.4]
fidList = map(add_a_function, weights)
assert isinstance(fidList, list)
assert len(fidList) == len(weights)
gm.addFactors(fidList, [[0, 1], [1, 2], [3, 4]])
def __init__(self,cgp): super(MulticutClustering, self).__init__(cgp) # build a graphical model nVar = cgp.numCells(2) nFac = cgp.numCells(1) cell1Bounds = cgp.cell1BoundsArray()-1 self.gm = opengm.gm(numpy.ones(nVar,dtype=opengm.label_type)*nVar) # init with zero potts functions fids = self.gm.addFunctions(opengm.pottsFunctions([nVar,nVar],numpy.zeros(nFac),numpy.zeros(nFac) )) # add factors self.gm.addFactors(fids,cell1Bounds) self.cgc = opengm.inference.Cgc(gm=self.gm,parameter=opengm.InfParam(planar=True))
def multicutFromCgp2(cgp,e0,e1,parameter=None): boundArray = cgp.cell1BoundsArray()-1 nVar = cgp.numCells(2) nFac = cgp.numCells(1) space = numpy.ones(nVar,dtype=opengm.label_type)*nVar gm = opengm.gm(space) #w = numpy.require(weights,dtype=opengm.value_type) pf=opengm.pottsFunctions([nVar,nVar],e0,e1) fids = gm.addFunctions(pf) gm.addFactors(fids,boundArray) cgc = opengm.inference.Cgc(gm=gm,parameter=parameter) return cgc,gm
def _inference(self, node_potentials, inter_potentials,
edges, alg, return_energy=False, return_margin=False,
init=None, **kwargs):
if node_potentials.shape[1]!=self.nNodeLabel:
raise ValueError("Node feature function parameters should match node label numbers.")
if inter_potentials.shape[0]!=self.nNodeLabel or inter_potentials.shape[1]!=self.nNodeLabel \
or inter_potentials.shape[2]!=self.nEdgeLabel:
raise ValueError("Interaction potential function parameters should match combination number of labels.")
nNodes = node_potentials.shape[0]
nEdges = edges.shape[0]
gm = opengm.gm(np.hstack([np.ones(nNodes, dtype=opengm.label_type)*self.nNodeLabel, \
np.ones(nEdges, dtype=opengm.label_type)*self.nEdgeLabel]))
gm.reserveFactors(nNodes+2*nEdges)
gm.reserveFunctions(nNodes+2*nEdges,'explicit')
unaries = -np.require(node_potentials, dtype=opengm.value_type)
fidUnaries = gm.addFunctions(unaries)
visUnaries = np.arange(nNodes, dtype=np.uint64)
gm.addFactors(fidUnaries, visUnaries)
inter_potentials=np.repeat(inter_potentials[np.newaxis, :, :, :],
edges.shape[0], axis=0)
highOrderFunctions = -np.require(inter_potentials, dtype=opengm.value_type)
fidHighOrder = gm.addFunctions(highOrderFunctions)
vidHighOrder = np.hstack([edges, np.arange(nNodes, nNodes+nEdges).reshape((nEdges,1))])
vidHighOrder = np.require(vidHighOrder, dtype=np.uint64)
gm.addFactors(fidHighOrder, vidHighOrder)
if alg == 'bp':
inference = opengm.inference.BeliefPropagation(gm)
elif alg == 'trw':
inference = opengm.inference.TreeReweightedBp(gm)
if init is not None:
inference.setStartingPoint(init)
inference.infer()
res = inference.arg().astype(np.int)
if return_margin:
node_marginals = inference.marginals(np.arange(nNodes,
dtype=opengm.index_type))
return node_marginals
if return_energy:
return res, gm.evaluate(res)
return res
def infer(self, pairwise_scores, unary_scores, k, mrf_type,
filenames=None, target_class=None, bcd_scores=None):
'''
Returns:
- argmin: a list of size k where each element is the label selected for the
corresponding node.
- t: runing time for inference
'''
import opengm
num_nodes = self.num_nodes
if mrf_type.endswith('energy'):
energy = -pairwise_scores.sum()
argmin = np.zeros((num_nodes,), dtype=np.int32)
t = 1
return argmin, t, energy
num_edges = self.get_num_edges()
num_labels = int(np.sqrt(pairwise_scores.shape[1]))
indices = self.get_indices()
gm = opengm.gm([num_labels]*num_nodes, operator='adder')
if unary_scores is not None:
assert unary_scores.shape[0] == k
assert unary_scores.shape[1] == num_labels
for i in range(num_nodes):
gm.addFactor(gm.addFunction(-unary_scores[i]), [i])
for i,e in enumerate(indices):
gm.addFactor(gm.addFunction(-pairwise_scores[i].reshape(
(num_labels, num_labels))), [e[0],e[1]])
if mrf_type.endswith('astar'):
inf = opengm.inference.AStar(gm, accumulator='minimizer')
elif mrf_type.endswith('trws'):
inf = opengm.inference.TrwsExternal(gm, accumulator='minimizer')
elif mrf_type.endswith('trbp'):
inf = opengm.inference.TreeReweightedBp(gm, accumulator='minimizer')
elif mrf_type.endswith('map'):
inf = opengm.inference.BeliefPropagation(gm, accumulator='minimizer')
else:
raise ValueError('Inference for mrf type {} is not implemented'.format(mrf_type))
#br_inf = opengm.inference.Bruteforce(gm, accumulator='minimizer')
#br_inf.infer()
t0 = time.time()
inf.infer()
argmin = inf.arg()
t1 = time.time() - t0
energy = inf.value()
return argmin, t1, energy
def instantiate_sentence(sentence_data):
# Get the number of tokens:
numVar = len(sentence_data)
# print 'numVar', numVar
# The domain of the variables is the len of th y_list
# Use the multiplier, the adder doesn't seem to work
gm=opengm.gm([len(y_list)]*numVar,'multiplier')
# gm=opengm.gm([len(y_list)]*numVar,'adder')
# print gm.numberOfVariables
# exit()
# Create the unary factors for the pgm
gm, feat_idxs = create_unaries(gm, sentence_data)
# print gm.factors()
# for fac in gm.factors():
# print fac
return gm
def forward(self, unary_pots):
""" Receive input tensor, return output tensor"""
self.save_for_backward(unary_pots)
print("In forward")
b, r, c, k = unary_pots.size()
if (False):
if torch.cuda.is_available():
unaries = unary_pots.cpu().numpy()
else:
unaries = unary_pots.numpy()
unaries = unaries.reshape([b * r * c, k])
numVar = r * c
gm = opengm.gm(np.ones(numVar, dtype=opengm.label_type) * k)
uf_id = gm.addFunctions(unaries)
potts = opengm.PottsFunction([k, k], 0.0, 0.4)
pf_id = gm.addFunction(potts)
vis = np.arange(0, numVar, dtype=np.uint64)
# add all unary factors at once
gm.addFactors(uf_id, vis)
# add pairwise factors
### Row Factors
for i in range(0, r):
for j in range(0, c - 1):
gm.addFactor(pf_id, [i * c + j, i * c + j + 1])
### Column Factors
for i in range(0, r - 1):
for j in range(c):
gm.addFactor(pf_id, [i * c + j, (i + 1) * c + j])
print("Graphical Model Constructed")
inf = opengm.inference.AlphaExpansionFusion(gm)
inf.infer()
labels = inf.arg()
return torch.from_numpy(np.asarray(labels).astype('float'))
else:
return torch.zeros(b, r, c)
def run_mc_opengm(segmentation, edges, energies):
n_seg = np.max(segmentation)
states = np.ones(n_seg)
gm = opengm.gm(states)
print "AAA"
"pairwise"
potts_shape = [ 2, 2]
potts = opengm.pottsFunctions(potts_shape,
np.array([0.0]),
np.array(energies)
)
print "AAA"
fids_p = gm.addFunctions(potts)
gm.addFactors(fids_p, edges)
gm_path = "/tmp/gm.h5"
opengm.saveGm(gm, gm_path)
print "AAA"
"parameters"
# wf = "(TTC)(MTC)(IC)(CC-IFD,TTC-I)" # default workflow
#wf = "(IC)(TTC-I,CC-I)" # propper workflow
# wf = "(TTC)(TTC,CC)" # lp relaxation
param = opengm.InfParam()#workflow=wf)
print "---inference---"
print " starting time:", time.strftime("%H:%M:%S"), ";", time.strftime("%d/%m/%Y")
print "..."
inf = opengm.inference.Multicut(gm, parameter=param)
inf.infer()
print " end time:", time.strftime("%H:%M:%S"), ";", time.strftime("%d/%m/%Y")
res_node = inf.arg()
res_edge = inf.getEdgeLabeling()
res_seg = inf.getSegmentation()
print res_node.shape, np.unique(res_node)
print res_edge.shape, np.unique(res_edge)
print res_seg.shape, np.unique(res_seg)
quit()
def generate_mc_grid(dimx, dimy, operator="adder"):
labels=dimx*dimy
nos = numpy.ones(labels, dtype=numpy.uint64) * labels
gm = opengm.gm(nos, operator, 0)
for y in range(dimy):
for x in range(dimx):
if x + 1 < dimx:
vis = [x + y * dimx, x + 1 + y * dimx]
assert vis.sort is not None
vis.sort
l=random.random()*2.0 - 1.0
fr=opengm.pottsFunction([labels,labels],0.0,l)
fid2=gm.addFunction(fr)
gm.addFactor(fid2, vis)
if y + 1 < dimy:
vis = [x + y * dimx, x + (y + 1) * dimx]
vis.sort()
l=random.random()*2.0 - 1.0
fr=opengm.pottsFunction([labels,labels],0.0,l)
fid2=gm.addFunction(fr)
gm.addFactor(fid2, vis)
return gm
def generate_mc_grid(dimx, dimy, operator="adder"):
labels=dimx*dimy
nos = numpy.ones(labels, dtype=numpy.uint64) * labels
gm = opengm.gm(nos, operator, 0)
for y in range(dimy):
for x in range(dimx):
if x + 1 < dimx:
vis = [x + y * dimx, x + 1 + y * dimx]
assert vis.sort is not None
vis.sort
l=random.random()*2.0 - 1.0
fr=opengm.pottsFunction([labels,labels],0.0,l)
fid2=gm.addFunction(fr)
gm.addFactor(fid2, vis)
if y + 1 < dimy:
vis = [x + y * dimx, x + (y + 1) * dimx]
vis.sort()
l=random.random()*2.0 - 1.0
fr=opengm.pottsFunction([labels,labels],0.0,l)
fid2=gm.addFunction(fr)
gm.addFactor(fid2, vis)
return gm
def test_add_factors_generic(self):
def mygen():
yield 0
yield 1
gm = opengm.gm([2, 4])
f = opengm.PottsFunction([2, 4], 0.0, 1.0)
fid = gm.addFunction(f)
vis_list = [
[0, 1],
(0, 1),
(x for x in xrange(2)),
mygen(),
opengm.IndexVector(x for x in xrange(0, 2)),
numpy.arange(0, 2, dtype=numpy.uint64)
]
for i, vis in enumerate(vis_list):
fIndex = gm.addFactor(fid, vis)
assert(gm.numberOfFactors == i + 1)
assert(fIndex == i)
assert(gm[fIndex].numberOfVariables == 2)
assert(gm[fIndex].shape[0] == 2)
assert(gm[fIndex].shape[1] == 4)
assert(gm[fIndex].variableIndices[0] == 0)
assert(gm[fIndex].variableIndices[1] == 1)
def onClickedMulticut(self):
p1 = self.probs.copy()
p1 = numpy.clip(p1, 0.005, 1.0 - 0.005)
p0 = 1.0 - p1
weights = numpy.log(p0 / p1)
nVar = self.rag.maxNodeId + 1
nos = numpy.ones(nVar) * nVar
gm = opengm.gm(nos)
uv = self.rag.uvIds()
uv = numpy.sort(uv, axis=1)
pf = opengm.pottsFunctions([nVar, nVar], numpy.array([0]), weights)
fid = gm.addFunctions(pf)
gm.addFactors(fid, uv)
inf = opengm.inference.Multicut(gm)
inf.infer(inf.verboseVisitor())
arg = inf.arg()
self.eArg = arg[uv[:, 0]] != arg[uv[:, 1]]
self.ctrlWidget.modeSelectorComboBox.setCurrentIndex(6)
def onClickedMulticut(self):
p1 = self.probs.copy()
p1 = numpy.clip(p1, 0.005, 1.0-0.005)
p0 = 1.0 - p1
weights = numpy.log(p0/p1)
nVar = self.rag.maxNodeId + 1
nos = numpy.ones(nVar)*nVar
gm = opengm.gm(nos)
uv = self.rag.uvIds()
uv = numpy.sort(uv,axis=1)
pf = opengm.pottsFunctions([nVar,nVar], numpy.array([0]),weights)
fid = gm.addFunctions(pf)
gm.addFactors(fid,uv)
inf = opengm.inference.Multicut(gm)
inf.infer(inf.verboseVisitor())
arg = inf.arg()
self.eArg = arg[uv[:,0]]!=arg[uv[:,1]]
self.ctrlWidget.modeSelectorComboBox.setCurrentIndex(6)
def denoiseModel(img,
norm=2,
weight=1.0,
truncate=None,
numLabels=256,
neighbourhood=4,
inpaintPixels=None,
randInpaitStartingPoint=False):
"""
this function is used to set up a graphical model similar to
**Denoising and inpainting problems:** from `Mrf- Benchmark <http://vision.middlebury.edu/MRF/results/ >`_
Args :
img : a grayscale image in the range [0,256)
norm : used norm for unaries and 2-order functions (default : 2)
weight : weight of 2-order functions (default : 1.0)
truncate : Truncate second order function at an given value (defaut : None)
numLabels : number of labels for each variable in the graphical model,
set this to a lower number to speed up inference (default : 255)
neighbourhood : neighbourhood for the second order functions, so far only 4 is allowed (default : 4)
inpaintPixels : a tuple of x and y coordinates where no unaries are added
randInpaitStartingPoint : use a random starting point for all pixels without unaries (default : False)
"""
shape = img.shape
if (img.ndim != 2):
raise RuntimeError("image must be gray")
if neighbourhood != 4:
raise RuntimeError(
"A neighbourhood other than 4 is not yet implemented")
# normalize and flatten image
iMin = numpy.min(img)
iMax = numpy.max(img)
imgNorm = ((img[:, :] - iMin) / (iMax - iMin)) * float(numLabels)
imgFlat = imgNorm.reshape(-1).astype(numpy.uint64)
# Set up Grapical Model:
numVar = int(img.size)
gm = opengm.gm([numLabels] * numVar, operator='adder')
gm.reserveFunctions(numLabels, 'explicit')
numberOfPairwiseFactors = shape[0] * (shape[1] -
1) + shape[1] * (shape[0] - 1)
gm.reserveFactors(numVar - len(inpaintPixels[0]) + numberOfPairwiseFactors)
# Set up unaries:
# - create a range of all possible labels
allPossiblePixelValues = numpy.arange(numLabels)
pixelValueRep = numpy.repeat(allPossiblePixelValues[:, numpy.newaxis],
numLabels, 1)
# - repeat [0,1,2,3,...,253,254,255] numVar times
labelRange = numpy.arange(numLabels, dtype=opengm.value_type)
labelRange = numpy.repeat(labelRange[numpy.newaxis, :], numLabels, 0)
unaries = numpy.abs(pixelValueRep - labelRange)**norm
# - add unaries to the graphical model
fids = gm.addFunctions(unaries.astype(opengm.value_type))
# add unary factors to graphical model
if (inpaintPixels is None):
for l in xrange(numLabels):
whereL = numpy.where(imgFlat == l)
gm.addFactors(fids[l], whereL[0].astype(opengm.index_type))
else:
# get vis of inpaint pixels
ipX = inpaintPixels[0]
ipY = inpaintPixels[1]
ipVi = ipX * shape[1] + ipY
for l in xrange(numLabels):
whereL = numpy.where(imgFlat == l)
notInInpaint = numpy.setdiff1d(whereL[0], ipVi)
gm.addFactors(fids[l], notInInpaint.astype(opengm.index_type))
# add ONE second order function
f = opengm.differenceFunction(shape=[numLabels, numLabels],
norm=2,
weight=weight)
fid = gm.addFunction(f)
vis2Order = opengm.secondOrderGridVis(shape[0], shape[1], True)
# add all second order factors
gm.addFactors(fid, vis2Order)
# create a starting point
startingPoint = imgFlat.copy()
if randInpaitStartingPoint:
startingPointRandom = numpy.random.randint(
0, numLabels, size=numVar).astype(opengm.index_type)
ipVi = inpaintPixels[0] * shape[1] + inpaintPixels[1]
for x in ipVi:
startingPoint[x] = startingPointRandom[x]
startingPoint[startingPoint == numLabels] = numLabels - 1
return gm, startingPoint.astype(opengm.index_type)
def crftest():
"""
pip install pyqpbo
pip install pystruct
http://taku910.github.io/crfpp/#install
cd ~/tmp
#wget https://drive.google.com/folderview?id=0B4y35FiV1wh7fngteFhHQUN2Y1B5eUJBNHZUemJYQV9VWlBUb3JlX0xBdWVZTWtSbVBneU0&usp=drive_web#list
7z x CRF++-0.58.tar.gz
7z x CRF++-0.58.tar
cd CRF++-0.58
chmod +x configure
./configure
make
"""
import pystruct
import pystruct.models
inference_method_options = ['lp', 'max-product']
inference_method = inference_method_options[1]
# graph = pystruct.models.GraphCRF(
# n_states=None,
# n_features=None,
# inference_method=inference_method,
# class_weight=None,
# directed=False,
# )
num_annots = 5
num_names = num_annots
aids = np.arange(5)
rng = np.random.RandomState(0)
hidden_nids = rng.randint(0, num_names, num_annots)
unique_nids, groupxs = ut.group_indices(hidden_nids)
# Indicator vector indicating the name
node_features = np.zeros((num_annots, num_names))
node_features[(aids, hidden_nids)] = 1
toy_params = {True: {'mu': 1.0, 'sigma': 2.2}, False: {'mu': 7.0, 'sigma': 0.9}}
if False:
import vtool as vt
import wbia.plottool as pt
pt.ensureqt()
xdata = np.linspace(0, 100, 1000)
tp_pdf = vt.gauss_func1d(xdata, **toy_params[True])
fp_pdf = vt.gauss_func1d(xdata, **toy_params[False])
pt.plot_probabilities([tp_pdf, fp_pdf], ['TP', 'TF'], xdata=xdata)
def metric(aidx1, aidx2, hidden_nids=hidden_nids, toy_params=toy_params):
if aidx1 == aidx2:
return 0
rng = np.random.RandomState(int(aidx1 + aidx2))
same = hidden_nids[int(aidx1)] == hidden_nids[int(aidx2)]
mu, sigma = ut.dict_take(toy_params[same], ['mu', 'sigma'])
return np.clip(rng.normal(mu, sigma), 0, np.inf)
pairwise_aidxs = list(ut.iprod(range(num_annots), range(num_annots)))
pairwise_labels = np.array( # NOQA
[hidden_nids[a1] == hidden_nids[a2] for a1, a2 in pairwise_aidxs]
)
pairwise_scores = np.array([metric(*zz) for zz in pairwise_aidxs])
pairwise_scores_mat = pairwise_scores.reshape(num_annots, num_annots) # NOQA
graph = pystruct.models.EdgeFeatureGraphCRF( # NOQA
n_states=num_annots,
n_features=num_names,
n_edge_features=1,
inference_method=inference_method,
)
import opengm
numVar = 10
unaries = np.ones([numVar, 3], dtype=opengm.value_type)
gm = opengm.gm(np.ones(numVar, dtype=opengm.label_type) * 3)
unary_fids = gm.addFunctions(unaries)
gm.addFactors(unary_fids, np.arange(numVar))
infParam = opengm.InfParam(workflow=ut.ensure_ascii('(IC)(TTC-I,CC-I)'))
inf = opengm.inference.Multicut(gm, parameter=infParam)
visitor = inf.verboseVisitor(printNth=1, multiline=False)
inf.infer(visitor)
arg = inf.arg()
# gridVariableIndices = opengm.secondOrderGridVis(img.shape[0], img.shape[1])
# fid = gm.addFunction(regularizer)
# gm.addFactors(fid, gridVariableIndices)
# regularizer = opengm.pottsFunction([3, 3], 0.0, beta)
# gridVariableIndices = opengm.secondOrderGridVis(img.shape[0], img.shape[1])
# fid = gm.addFunction(regularizer)
# gm.addFactors(fid, gridVariableIndices)
unaries = np.random.rand(10, 10, 2)
potts = opengm.PottsFunction([2, 2], 0.0, 0.4)
gm = opengm.grid2d2Order(unaries=unaries, regularizer=potts)
inf = opengm.inference.GraphCut(gm)
inf.infer()
arg = inf.arg() # NOQA
def denoiseModel(
img,
norm = 2,
weight = 1.0,
truncate = None,
numLabels = 256,
neighbourhood = 4,
inpaintPixels = None,
randInpaitStartingPoint = False
):
"""
this function is used to set up a graphical model similar to
**Denoising and inpainting problems:** from `Mrf- Benchmark <http://vision.middlebury.edu/MRF/results/ >`_
Args :
img : a grayscale image in the range [0,256)
norm : used norm for unaries and 2-order functions (default : 2)
weight : weight of 2-order functions (default : 1.0)
truncate : Truncate second order function at an given value (defaut : None)
numLabels : number of labels for each variable in the graphical model,
set this to a lower number to speed up inference (default : 255)
neighbourhood : neighbourhood for the second order functions, so far only 4 is allowed (default : 4)
inpaintPixels : a tuple of x and y coordinates where no unaries are added
randInpaitStartingPoint : use a random starting point for all pixels without unaries (default : False)
"""
shape = img.shape
if(img.ndim!=2):
raise RuntimeError("image must be gray")
if neighbourhood != 4 :
raise RuntimeError("A neighbourhood other than 4 is not yet implemented")
# normalize and flatten image
iMin = numpy.min(img)
iMax = numpy.max(img)
imgNorm = ((img[:,:]-iMin)/(iMax-iMin))*float(numLabels)
imgFlat = imgNorm.reshape(-1).astype(numpy.uint64)
# Set up Grapical Model:
numVar = int(img.size)
gm = opengm.gm([numLabels]*numVar,operator='adder')
gm.reserveFunctions(numLabels,'explicit')
numberOfPairwiseFactors=shape[0]*(shape[1]-1) + shape[1]*(shape[0]-1)
gm.reserveFactors(numVar-len(inpaintPixels[0]) + numberOfPairwiseFactors )
# Set up unaries:
# - create a range of all possible labels
allPossiblePixelValues=numpy.arange(numLabels)
pixelValueRep = numpy.repeat(allPossiblePixelValues[:,numpy.newaxis],numLabels,1)
# - repeat [0,1,2,3,...,253,254,255] numVar times
labelRange = numpy.arange(numLabels,dtype=opengm.value_type)
labelRange = numpy.repeat(labelRange[numpy.newaxis,:], numLabels, 0)
unaries = numpy.abs(pixelValueRep - labelRange)**norm
# - add unaries to the graphical model
fids=gm.addFunctions(unaries.astype(opengm.value_type))
# add unary factors to graphical model
if(inpaintPixels is None):
for l in xrange(numLabels):
whereL=numpy.where(imgFlat==l)
gm.addFactors(fids[l],whereL[0].astype(opengm.index_type))
else:
# get vis of inpaint pixels
ipX = inpaintPixels[0]
ipY = inpaintPixels[1]
ipVi = ipX*shape[1] + ipY
for l in xrange(numLabels):
whereL=numpy.where(imgFlat==l)
notInInpaint=numpy.setdiff1d(whereL[0],ipVi)
gm.addFactors(fids[l],notInInpaint.astype(opengm.index_type))
# add ONE second order function
f=opengm.differenceFunction(shape=[numLabels,numLabels],norm=2,weight=weight)
fid=gm.addFunction(f)
vis2Order=opengm.secondOrderGridVis(shape[0],shape[1],True)
# add all second order factors
gm.addFactors(fid,vis2Order)
# create a starting point
startingPoint = imgFlat.copy()
if randInpaitStartingPoint :
startingPointRandom = numpy.random.randint(0,numLabels,size=numVar).astype(opengm.index_type)
ipVi = inpaintPixels[0]*shape[1] + inpaintPixels[1]
for x in ipVi:
startingPoint[x]=startingPointRandom[x]
startingPoint[startingPoint==numLabels]=numLabels-1
return gm,startingPoint.astype(opengm.index_type)
from opengm import learning np = numpy numLabels = 3 numVar = 6 ################################################################# # add a unary function ################################################################## print opengm.learning.DatasetWithHammingLoss print opengm.learning.HammingLoss # make the gm space = numpy.ones(numVar) * numLabels gm = opengm.gm(space) weightVals = numpy.ones(100) * 1.0 weights = opengm.learning.Weights(weightVals) ################################################################## # add a unary function ################################################################## features = numpy.ones([numLabels, 2], dtype=opengm.value_type) weightIds = numpy.ones([numLabels, 2], dtype=opengm.index_type) # set up weight ids for each label weightIds[0, :] = [0, 1] weightIds[1, :] = [2, 3] weightIds[2, :] = [4, 5]
def __init__(self,
tl_line_rec,
region_det,
line_rec,
line_pts,
stats,
scale=1.0,
sign_hypos=None,
param_dict=None):
# create graphical model from fragement
self.stats = stats
self.scale = scale
self.scaled_sign_height = stats.tblSignHeight * scale
self.min_sign_dist = self.scaled_sign_height / 2. # distance between sign centers
self.tl_line_rec = tl_line_rec
# null hypothesis for signs in tl
self.sign_hypos = sign_hypos
# detections contained in rectangluar area around respective alignments
# [ID, cx, cy, score, x1, y1, x2, y2, idx]
self.region_det = region_det
# init
self.num_vars = len(self.tl_line_rec)
self.num_relevant = 0
self.max_cost = 1e10 # 1e11 # "inifinite" cost
# only continue, if there is a sign in line to match
if self.num_vars > 0:
# compute num_lbls_per_var from detections
ulbls, counts = np.unique(self.region_det[:, 0],
return_counts=True)
hypo_det_counts = np.array([
counts[ulbls == item] if item in ulbls else 0
for item in self.tl_line_rec.lbl
],
dtype=int).squeeze()
self.tl_line_rec['det_count'] = hypo_det_counts
# optional: remove vars without detections
if False:
# self.tl_line_rec = self.tl_line_rec[hypo_det_counts > 0]
self.tl_line_rec = self.tl_line_rec.iloc[np.where(
hypo_det_counts > 0
)] # deal with scalar case of boolean indexing
# hypo_det_counts = hypo_det_counts[hypo_det_counts > 0]
# only continue, at least a single matching detection
self.num_relevant = np.sum(counts[np.isin(ulbls,
self.tl_line_rec.lbl)])
if self.num_relevant > 0:
# update num_vars
self.num_vars = len(self.tl_line_rec)
# opengm setup
self.num_lbls_per_var = max(counts[np.isin(
ulbls, self.tl_line_rec.lbl)]) + 1 # + 1 outlier detection
var_space = np.ones(self.num_vars) * self.num_lbls_per_var
self.gm = opengm.gm(var_space)
# parameter setup
if param_dict is not None:
self.params = param_dict
else:
self.params = dict()
# extra settings
self.params['outlier_cost'] = 10
self.params['angle_long_range'] = True
# unary potentials
self.params['lambda_score'] = 0.3
self.params['sigma_score'] = 0.4
self.params[
'lambda_offset'] = 1 # currently offset used linearly without exp function
self.params[
'sigma_offset'] = 1 # lambda & sigma have no influence!
# pairwise binary potentials
self.params['lambda_p'] = 3 # 1
self.params['sigma_p'] = 3
self.params['lambda_angle'] = 2
self.params['sigma_angle'] = 0.6
self.params['lambda_iou'] = 2
self.params['sigma_iou'] = 0.4
# OPTIONAL: strong penalties for long range connections
if True:
self.params['lr_lambda_angle'] = 0.05
self.params['lr_sigma_angle'] = 0.1
self.params['lr_lambda_iou'] = 0.1
self.params['lr_sigma_iou'] = 0.05
else:
self.params['lr_lambda_angle'] = self.params[
'lambda_angle']
self.params['lr_sigma_angle'] = self.params[
'sigma_angle']
self.params['lr_lambda_iou'] = self.params[
'lambda_iou']
self.params['lr_sigma_iou'] = self.params['sigma_iou']
# angle of hypothesis line
self.b = line_pts[-1, :] - line_pts[0, :]
# print 'hypo angle:', np.arctan2(self.b[1], self.b[0]) * (180 / np.pi), self.b
# offset
self.Xb = line_pts[0, :].reshape(1, -1)
# define variance between line distance and sign distance - for seculidean and mahalanobis
self.variance_p = np.array([1, 0.2],
dtype=np.float) # [8, 1] [1, 1]
if False:
print('#syms:', len(self.tl_line_rec), 'max#dets_per_sym:',
self.num_lbls_per_var - 1, 'relevant#dets:',
self.num_relevant, 'total#dets:',
self.region_det.shape[0])
# print(np.vstack([self.fm_hypo_df.lbl, hypo_det_counts])).astype(int)
# print self.fm_hypo_df
# assemble potentials
self.add_unary()
self.add_pairwise()
import opengm import numpy #------------------------------------------------------------------------------------ # This example shows how multiple unaries functions and functions / factors add once #------------------------------------------------------------------------------------ # add unaries from a for a 2d grid / image width=10 height=20 numVar=width*height numLabels=2 # construct gm gm=opengm.gm(numpy.ones(numVar,dtype=opengm.index_type)*numLabels) # construct an array with all unaries (random in this example) unaries=numpy.random.rand(width,height,numLabels) # reshape unaries is such way, that the first axis is for the different functions unaries2d=unaries.reshape([numVar,numLabels]) # add all unary functions at once (#numVar unaries) fids=gm.addFunctions(unaries2d) # numpy array with the variable indices for all factors vis=numpy.arange(0,numVar,dtype=numpy.uint64) # add all unary factors at once gm.addFactors(fids,vis)
import opengm import numpy #------------------------------------------------------------------------------------ # This example shows how multiple unaries functions and functions / factors add once #------------------------------------------------------------------------------------ # add unaries from a for a 2d grid / image width = 10 height = 20 numVar = width * height numLabels = 2 # construct gm gm = opengm.gm(numpy.ones(numVar, dtype=opengm.index_type) * numLabels) # construct an array with all unaries (random in this example) unaries = numpy.random.rand(width, height, numLabels) # reshape unaries is such way, that the first axis is for the different functions unaries2d = unaries.reshape([numVar, numLabels]) # add all unary functions at once (#numVar unaries) fids = gm.addFunctions(unaries2d) # numpy array with the variable indices for all factors vis = numpy.arange(0, numVar, dtype=numpy.uint64) # add all unary factors at once gm.addFactors(fids, vis)
import numpy
import opengm
import matplotlib.pyplot as plt
f1 = numpy.ones([2])
f2 = numpy.ones([2, 2])
f3 = numpy.ones([2, 2, 2])
"""
Triangle (non-shared) :
- 3 variables
- 3 unaries
- 2 second order functions
- 1 third order factor
- functions are *non* - shared
"""
gm = opengm.gm([2, 2, 2])
gm.addFactor(gm.addFunction(f1), [0])
gm.addFactor(gm.addFunction(f1), [1])
gm.addFactor(gm.addFunction(f1), [2])
gm.addFactor(gm.addFunction(f2), [0, 1])
gm.addFactor(gm.addFunction(f2), [1, 2])
gm.addFactor(gm.addFunction(f2), [0, 2])
gm.addFactor(gm.addFunction(f3), [0, 1, 2])
opengm.visualizeGm(gm, show=False, plotFunctions=True, plotNonShared=True)
plt.savefig("triangle.png", bbox_inches='tight', dpi=300)
plt.close()
import opengm
import numpy
gm=opengm.gm([2,2,3,3,4,4,4],operator='adder')
functionIds=[]
#---------------------------------------------------------------
# Numpy Ndarray
# (is stored in a different multi array function within opengm)
#---------------------------------------------------------------
f=numpy.random.rand(2,2,3,4)
fid=gm.addFunction(f)
gm.addFactor(fid,[0,1,2,4])
print "\nexplicit function: \n",f
#---------------------------------------------------------------
# Sparse Function
#---------------------------------------------------------------
# fill sparse function "by hand"
f=opengm.SparseFunction(shape=[3,4,4],defaultValue=1)
# fill diagonale with zeros
for d in xrange(4):
f[[d,d,d]]=0
print "\nsparse function: \n",f
fid=gm.addFunction(f)
functionIds.append(fid)
gm.addFactor(fid,[3,4,5])
# fill sparse function from dense function
img += noise
img -= img.min()
img /= img.max()
print "shape", img.shape
vigra.imshow(img)
#vigra.show()
threshold = 0.24
labelsNaive = img > threshold
vigra.imshow(labelsNaive)
#vigra.show()
nVar = img.size
nLabelsPerVar = 2
variableSpace = numpy.ones(nVar) * nLabelsPerVar
gm = opengm.gm(variableSpace)
t0 = time.time()
# add unaries
for y in range(img.shape[1]):
for x in range(img.shape[0]):
energy0 = img[x, y] - threshold
energy1 = threshold - img[x, y]
unaryFunction = numpy.array([energy0, energy1])
# add unary function to graphical model
functionId = gm.addFunction(unaryFunction)
# add unary factor to graphical model
variableIndex = y + x * img.shape[1]
def _inference(self,
node_potentials,
edge_potentials,
inter_potentials,
edges,
alg,
return_energy=False,
return_margin=False,
init=None,
**kwargs):
if node_potentials.shape[1] != self.nNodeLabel:
raise ValueError(
"Node feature function parameters should match node label numbers."
)
if edge_potentials.shape[0] != edges.shape[0]:
raise ValueError(
"Edge feature function numbers should match given edges.")
if edge_potentials.shape[1] != self.nEdgeLabel:
raise ValueError(
"Edge feature function parameters should match edge label numbers."
)
if inter_potentials.shape[0] != edges.shape[0]:
raise ValueError(
"Interaction potential function number should match edge numbers."
)
if inter_potentials.shape[1]!=self.nNodeLabel or inter_potentials.shape[2]!=self.nNodeLabel \
or inter_potentials.shape[3]!=self.nEdgeLabel:
raise ValueError(
"Interaction potential function parameters should match combination number of labels."
)
nNodes = node_potentials.shape[0]
nEdges = edges.shape[0]
gm = opengm.gm(np.hstack([np.ones(nNodes, dtype=opengm.label_type)*self.nNodeLabel, \
np.ones(nEdges, dtype=opengm.label_type)*self.nEdgeLabel]))
gm.reserveFactors(nNodes + 2 * nEdges)
gm.reserveFunctions(nNodes + 2 * nEdges, 'explicit')
unaries = -np.require(node_potentials, dtype=opengm.value_type)
fidUnaries = gm.addFunctions(unaries)
visUnaries = np.arange(nNodes, dtype=np.uint64)
gm.addFactors(fidUnaries, visUnaries)
unaries = -np.require(edge_potentials, dtype=opengm.value_type)
fidUnaries = gm.addFunctions(unaries)
visUnaries = np.arange(nNodes, nEdges + nNodes, dtype=np.uint64)
gm.addFactors(fidUnaries, visUnaries)
highOrderFunctions = -np.require(inter_potentials,
dtype=opengm.value_type)
fidHighOrder = gm.addFunctions(highOrderFunctions)
vidHighOrder = np.hstack(
[edges,
np.arange(nNodes, nNodes + nEdges).reshape((nEdges, 1))])
vidHighOrder = np.require(vidHighOrder, dtype=np.uint64)
gm.addFactors(fidHighOrder, vidHighOrder)
if alg == 'bp':
inference = opengm.inference.BeliefPropagation(gm)
elif alg == 'dd':
inference = opengm.inference.DualDecompositionSubgradient(gm)
elif alg == 'trws':
inference = opengm.inference.TrwsExternal(gm)
elif alg == 'trw':
inference = opengm.inference.TreeReweightedBp(gm)
elif alg == 'gibbs':
inference = opengm.inference.Gibbs(gm)
elif alg == 'lf':
inference = opengm.inference.LazyFlipper(gm)
elif alg == 'icm':
inference = opengm.inference.Icm(gm)
elif alg == 'dyn':
inference = opengm.inference.DynamicProgramming(gm)
elif alg == 'fm':
inference = opengm.inference.AlphaExpansionFusion(gm)
elif alg == 'gc':
inference = opengm.inference.GraphCut(gm)
elif alg == 'loc':
inference = opengm.inference.Loc(gm)
elif alg == 'mqpbo':
inference = opengm.inference.Mqpbo(gm)
elif alg == 'alphaexp':
inference = opengm.inference.AlphaExpansion(gm)
elif alg == 'lp':
parameter = opengm.InfParam(integerConstraint=True)
inference = opengm.inference.LpCplex(gm, parameter=parameter)
if init is not None:
inference.setStartingPoint(init)
inference.infer()
res = inference.arg().astype(np.int)
if return_margin:
node_marginals = inference.marginals(
np.arange(nNodes, dtype=opengm.index_type))
edge_marginals = inference.marginals(
np.arange(nNodes, nNodes + nEdges, dtype=opengm.index_type))
return (node_marginals, edge_marginals)
if return_energy:
return res, gm.evaluate(res)
return res
def _inference(self,
node_potentials,
inter_potentials,
edges,
alg,
return_energy=False,
return_margin=False,
init=None,
**kwargs):
if node_potentials.shape[1] != self.nNodeLabel:
raise ValueError(
"Node feature function parameters should match node label numbers."
)
if inter_potentials.shape[0]!=self.nNodeLabel or inter_potentials.shape[1]!=self.nNodeLabel \
or inter_potentials.shape[2]!=self.nEdgeLabel:
raise ValueError(
"Interaction potential function parameters should match combination number of labels."
)
nNodes = node_potentials.shape[0]
nEdges = edges.shape[0]
gm = opengm.gm(np.hstack([np.ones(nNodes, dtype=opengm.label_type)*self.nNodeLabel, \
np.ones(nEdges, dtype=opengm.label_type)*self.nEdgeLabel]))
gm.reserveFactors(nNodes + 2 * nEdges)
gm.reserveFunctions(nNodes + 2 * nEdges, 'explicit')
unaries = -np.require(node_potentials, dtype=opengm.value_type)
fidUnaries = gm.addFunctions(unaries)
visUnaries = np.arange(nNodes, dtype=np.uint64)
gm.addFactors(fidUnaries, visUnaries)
inter_potentials = np.repeat(inter_potentials[np.newaxis, :, :, :],
edges.shape[0],
axis=0)
highOrderFunctions = -np.require(inter_potentials,
dtype=opengm.value_type)
fidHighOrder = gm.addFunctions(highOrderFunctions)
vidHighOrder = np.hstack(
[edges,
np.arange(nNodes, nNodes + nEdges).reshape((nEdges, 1))])
vidHighOrder = np.require(vidHighOrder, dtype=np.uint64)
gm.addFactors(fidHighOrder, vidHighOrder)
if alg == 'bp':
inference = opengm.inference.BeliefPropagation(gm)
elif alg == 'trw':
inference = opengm.inference.TreeReweightedBp(gm)
if init is not None:
inference.setStartingPoint(init)
inference.infer()
res = inference.arg().astype(np.int)
if return_margin:
node_marginals = inference.marginals(
np.arange(nNodes, dtype=opengm.index_type))
return node_marginals
if return_energy:
return res, gm.evaluate(res)
return res
import matplotlib.pyplot as plt
f1=numpy.ones([2])
f2=numpy.ones([2,2])
"""
Grid:
- 4x4=16 variables
- second order factors in 4-neigbourhood
all connected to the same function
- higher order functions are shared
"""
size=3
gm=opengm.gm([2]*size*size)
fid=gm.addFunction(f2)
for y in range(size):
for x in range(size):
gm.addFactor(gm.addFunction(f1),x*size+y)
if(x+1<size):
gm.addFactor(fid,[x*size+y,(x+1)*size+y])
if(y+1<size):
gm.addFactor(fid,[x*size+y,x*size+(y+1)])
opengm.visualizeGm( gm,layout='spring',iterations=3000,
show=False,plotFunctions=True,
plotNonShared=True,relNodeSize=0.4)
plt.savefig("grid.png",bbox_inches='tight',dpi=300)
def generate_pgm(if_data, verbose=False):
# gather data from the if data object
query_graph = if_data.current_sg_query
object_detections = if_data.object_detections
attribute_detections = if_data.attribute_detections
relationship_models = if_data.relationship_models
per_object_attributes = if_data.per_object_attributes
image_filename = if_data.image_filename
# generate the graphical model (vg_data_build_gm_for_image)
n_objects = len(query_graph.objects)
n_vars = []
object_is_detected = []
query_to_pgm = []
pgm_to_query = []
master_box_coords = []
varcount = 0
for obj_ix in range(0, n_objects):
query_object_name = query_graph.objects[obj_ix].names
# occasionally, there are multiple object names (is 0 the best?)
if isinstance(query_object_name, np.ndarray):
query_object_name = query_object_name[0]
object_name = "obj:" + query_object_name
if object_name not in object_detections:
object_is_detected.append(False)
query_to_pgm.append(-1)
else:
if len(master_box_coords) == 0:
master_box_coords = np.copy(object_detections[object_name][:,0:4])
object_is_detected.append(True)
query_to_pgm.append(varcount)
varcount += 1
pgm_to_query.append(obj_ix)
n_labels = len(object_detections[object_name])
n_vars.append(n_labels)
gm = ogm.gm(n_vars, operator='adder')
functions = []
# generate 1st order functions for objects
# TODO: test an uniform dist for missing objects
unary_dets = []
is_cnn_detected = []
for obj_ix in range(0, n_objects):
fid = None
pgm_ix = query_to_pgm[obj_ix]
object_name = query_graph.objects[obj_ix].names
if isinstance(object_name, np.ndarray):
object_name = object_name[0]
detail = "unary function for object '{0}'".format(object_name)
if object_is_detected[obj_ix]:
is_cnn_detected.append(True)
prefix_object_name = "obj:" + object_name
detections = object_detections[prefix_object_name]
unary_dets.append(detections[:,4])
log_scores = -np.log(detections[:,4])
fid = gm.addFunction(log_scores)
else:
continue
func_detail = FuncDetail(fid, [pgm_ix], "explicit", "object unaries", detail)
functions.append(func_detail)
#generate 1st order functions for attributes
n_attributes = len(per_object_attributes)
for attr_ix in range(0, n_attributes):
obj_ix = int(per_object_attributes[attr_ix][0])
pgm_ix = query_to_pgm[obj_ix]
attribute_name = per_object_attributes[attr_ix][1]
prefix_attribute_name = "atr:" + attribute_name
if prefix_attribute_name not in attribute_detections:
continue
if not object_is_detected[obj_ix]:
continue
detections = attribute_detections[prefix_attribute_name]
log_scores = -np.log(detections[:,4])
detail = "unary function for attribute '{0}' of object '{1}' (qry_ix:{2}, pgm_ix:{3})".format(attribute_name, query_graph.objects[obj_ix].names, obj_ix, pgm_ix)
fid = gm.addFunction(log_scores)
func_detail = FuncDetail(fid, [pgm_ix], "explicit", "attribute unaries", detail)
functions.append(func_detail)
# generate a tracker for storing obj/attr/rel likelihoods (pre-inference)
tracker = DetectionTracker(image_filename)
for i in range(0, n_objects):
if object_is_detected[i]:
if isinstance(query_graph.objects[i].names, np.ndarray):
tracker.object_names.append(query_graph.objects[i].names[0])
else:
tracker.object_names.append(query_graph.objects[i].names)
tracker.unary_detections = unary_dets
tracker.box_coords = master_box_coords
tracker.detected_vars = is_cnn_detected
# generate 2nd order functions for binary relationships
trip_root = query_graph.binary_triples
trip_list = []
if isinstance(trip_root, sio.matlab.mio5_params.mat_struct):
trip_list.append(query_graph.binary_triples)
else:
# if there's only one relationship, we don't have an array :/
for trip in trip_root:
trip_list.append(trip)
# generate a single cartesian product of the boxes
# this will only work when all objects are detected across the same boxes
# we know this is the case for this implementation
master_cart_prod = None
for i in range(0, n_objects):
if object_is_detected[i]:
obj_name = query_graph.objects[i].names
boxes = None
if isinstance(obj_name, np.ndarray):
boxes = object_detections["obj:"+obj_name[0]]
else:
boxes = object_detections["obj:"+obj_name]
master_cart_prod = np.array([x for x in itertools.product(boxes, boxes)])
break
tracker.box_pairs = master_cart_prod
# process each binary triple in the list
for trip in trip_list:
sub_ix = trip.subject
sub_pgm_ix = query_to_pgm[sub_ix]
subject_name = query_graph.objects[sub_ix].names
if isinstance(subject_name, np.ndarray):
subject_name = subject_name[0]
obj_ix = trip.object
obj_pgm_ix = query_to_pgm[obj_ix]
object_name = query_graph.objects[obj_ix].names
if isinstance(object_name, np.ndarray):
object_name = object_name[0]
relationship = trip.predicate
bin_trip_key = subject_name + "_" + relationship.replace(" ", "_") + "_" + object_name
# check if there is a gmm for the specific triple string
if bin_trip_key not in relationship_models:
bin_trip_key = "*_" + relationship.replace(" ", "_") + "_*"
if bin_trip_key not in relationship_models:
continue
# verify object detections
if sub_ix == obj_ix:
continue
if not object_is_detected[sub_ix]:
continue
if not object_is_detected[obj_ix]:
continue
# get model parameters
prefix_object_name = "obj:" + object_name
bin_object_box = object_detections[prefix_object_name]
prefix_subject_name = "obj:" + subject_name
bin_subject_box = object_detections[prefix_subject_name]
rel_params = relationship_models[bin_trip_key]
# generate features from subject and object detection boxes
cart_prod = master_cart_prod
sub_dim = 0
obj_dim = 1
subx_center = cart_prod[:, sub_dim, 0] + 0.5 * cart_prod[:, sub_dim, 2]
suby_center = cart_prod[:, sub_dim, 1] + 0.5 * cart_prod[:, sub_dim, 3]
objx_center = cart_prod[:, obj_dim, 0] + 0.5 * cart_prod[:, obj_dim, 2]
objy_center = cart_prod[:, obj_dim, 1] + 0.5 * cart_prod[:, obj_dim, 3]
sub_width = cart_prod[:, sub_dim, 2]
relx_center = (subx_center - objx_center) / sub_width
sub_height = cart_prod[:, sub_dim, 3]
rely_center = (suby_center - objy_center) / sub_height
rel_height = cart_prod[:, obj_dim, 2] / cart_prod[:, sub_dim, 2]
rel_width = cart_prod[:, obj_dim, 3] / cart_prod[:, sub_dim, 3]
features = np.vstack((relx_center, rely_center, rel_height, rel_width)).T
#tracker.box_pairs = np.copy(cart_prod) #TODO: is this copy necessary?
#tracker.box_pairs = cart_prod
# generate scores => log(epsilon+scores) => platt sigmoid
scores = gmm_pdf(features, rel_params.gmm_weights, rel_params.gmm_mu, rel_params.gmm_sigma)
eps = np.finfo(np.float).eps
scores = np.log(eps + scores)
sig_scores = 1.0 / (1. + np.exp(rel_params.platt_a * scores + rel_params.platt_b))
log_likelihoods = -np.log(sig_scores)
#tracker.add_group(bin_trip_key, np.copy(log_likelihoods), np.copy(bin_object_box), object_name, np.copy(bin_subject_box), subject_name) # TODO: are these copy calls necessary?
tracker.add_group(bin_trip_key, log_likelihoods, bin_object_box, object_name, obj_pgm_ix, bin_subject_box, subject_name, sub_pgm_ix)
# generate the matrix of functions
n_subject_val = len(bin_subject_box)
n_object_val = len(bin_object_box)
bin_functions = np.reshape(log_likelihoods, (n_subject_val, n_object_val)) # TODO: determine if any transpose is needed
if obj_pgm_ix < sub_pgm_ix: bin_functions = bin_functions.T
# add binary functions to the GM
detail = "binary functions for relationship '%s'" % (bin_trip_key)
fid = gm.addFunction(bin_functions)
var_indices = [sub_pgm_ix, obj_pgm_ix]
if obj_pgm_ix < sub_pgm_ix: var_indices = [obj_pgm_ix, sub_pgm_ix]
func_detail = FuncDetail(fid, var_indices, "explicit", "binary functions", detail)
functions.append(func_detail)
# add 1st order factors (fid)
for f in functions:
n_indices = len(f.var_indices)
if n_indices == 1:
gm.addFactor(f.gm_function_id, f.var_indices[0])
elif n_indices == 2:
gm.addFactor(f.gm_function_id, f.var_indices)
else:
continue
return gm, tracker
def dummy_cut_example():
r"""
CommandLine:
python -m ibeis.workflow --exec-dummy_cut_example --show
Example:
>>> # DISABLE_DOCTEST
>>> from ibeis.workflow import * # NOQA
>>> result = dummy_cut_example()
>>> print(result)
>>> ut.quit_if_noshow()
>>> import plottool as pt
>>> ut.show_if_requested()
"""
import opengm
import numpy as np
import plottool as pt
pt.ensure_pylab_qt4()
# Matching Graph
cost_matrix = np.array([
[0.5, 0.6, 0.2, 0.4],
[0.0, 0.5, 0.2, 0.9],
[0.0, 0.0, 0.5, 0.1],
[0.0, 0.0, 0.0, 0.5],
])
cost_matrix += cost_matrix.T
number_of_labels = 4
num_annots = 4
#cost_matrix = (cost_matrix * 2) - 1
#gm = opengm.gm(number_of_labels)
gm = opengm.gm(np.ones(num_annots) * number_of_labels)
aids = np.arange(num_annots)
aid_pairs = np.array([(a1, a2) for a1, a2 in ut.iprod(
aids, aids) if a1 != a2], dtype=np.uint32)
aid_pairs.sort(axis=1)
# add a potts function
# penalizes neighbors for having different labels
# beta = 0 # 0.1 # strength of potts regularizer
#beta = 0.1 # 0.1 # strength of potts regularizer
# Places to look for the definition of this stupid class
# ~/code/opengm/src/interfaces/python/opengm/opengmcore/pyFunctionTypes.cxx
# /src/interfaces/python/opengm/opengmcore/function_injector.py
#shape = [number_of_labels] * 2
#regularizer = opengm.PottsGFunction(shape, 0.0, beta)
# __init__( (object)arg1, (object)shape [, (object)values=()]) -> object :
# values = np.arange(1, ut.num_partitions(num_annots) + 1)
#regularizer = opengm.PottsGFunction(shape)
#reg_fid = gm.addFunction(regularizer)
# A Comparative Study of Modern Inference Techniques for Structured Discrete Energy Minimization Problems
# http://arxiv.org/pdf/1404.0533.pdf
# regularizer1 = opengm.pottsFunction([number_of_labels] * 2, valueEqual=0.0, valueNotEqual=beta)
# gm.addFactors(reg_fid, aid_pairs)
# 2nd order function
pair_fid = gm.addFunction(cost_matrix)
gm.addFactors(pair_fid, aid_pairs)
if False:
Inf = opengm.inference.BeliefPropagation
parameter = opengm.InfParam(steps=10, damping=0.5, convergenceBound=0.001)
else:
Inf = opengm.inference.Multicut
parameter = opengm.InfParam()
inf = Inf(gm, parameter=parameter)
class PyCallback(object):
def __init__(self,):
self.labels = []
def begin(self, inference):
print("begin of inference")
def end(self, inference):
self.labels.append(inference.arg())
def visit(self, inference):
gm = inference.gm()
labelVector = inference.arg()
print("energy %r" % (gm.evaluate(labelVector),))
self.labels.append(labelVector)
callback = PyCallback()
visitor = inf.pythonVisitor(callback, visitNth=1)
inf.infer(visitor)
print(callback.labels)
print(cost_matrix)
pt.imshow(cost_matrix, cmap='magma')
opengm.visualizeGm(gm=gm)
pass
vigra.filters.gaussianGradientMagnitude(randGt.astype('float32'),
sigma))
a = None
b = None
if len(feat) > 0:
a = numpy.rollaxis(numpy.array(feat), axis=0, start=3)
if len(featB) > 0:
b = numpy.rollaxis(numpy.array(featB), axis=0, start=3)
return a, b
for mi in range(nModels):
#print mi
gm = opengm.gm(numpy.ones(numVar) * nLables)
gt = makeGt(shape)
gtFlat = gt.reshape([-1])
unaries, binaries = makeFeatures(gt)
# print unaries, binaries
for x in range(shape[0]):
for y in range(shape[1]):
uFeat = numpy.append(unaries[x, y, :], [1])
#print uFeat
#print uWeightIds
#print(unaries[x,y,:])
#print(unaries.shape)
0], " infile hdf5-outfile red green blue T lambda"
sys.exit(0)
img = vigra.readImage(args[1])
if img.shape[2] != 3:
print "Image must be RGB"
sys.exit(0)
T = float(args[6])
beta = float(args[7])
imgFlat = img.reshape([-1, 3]).view(numpy.ndarray)
numVar = imgFlat.shape[0]
gm = opengm.gm(numpy.ones(numVar, dtype=opengm.label_type) * 2)
protoColor = numpy.array([args[3], args[4], args[5]],
dtype=opengm.value_type).reshape([3, -1])
protoColor = numpy.repeat(protoColor, numVar, axis=1).swapaxes(0, 1)
diffArray = numpy.sum(numpy.abs(imgFlat - protoColor), axis=1)
unaries = numpy.ones([numVar, 2], dtype=opengm.value_type)
unaries[:, 0] = T
unaries[:, 1] = diffArray
print diffArray
gm.addFactors(gm.addFunctions(unaries), numpy.arange(numVar))
regularizer = opengm.pottsFunction([2, 2], 0.0, beta)
gridVariableIndices = opengm.secondOrderGridVis(img.shape[0], img.shape[1])
def inference_ogm(unary_potentials,
pairwise_potentials,
edges,
return_energy=False,
alg='dd',
init=None,
reserveNumFactorsPerVariable=2,
**kwargs):
"""Inference with OpenGM backend.
Parameters
----------
unary_potentials : nd-array, shape (n_nodes, n_nodes)
Unary potentials of energy function.
pairwise_potentials : nd-array, shape (n_states, n_states) or (n_states, n_states, n_edges).
Pairwise potentials of energy function.
If the first case, edge potentials are assumed to be the same for all edges.
In the second case, the sequence needs to correspond to the edges.
edges : nd-array, shape (n_edges, 2)
Graph edges for pairwise potentials, given as pair of node indices. As
pairwise potentials are not assumed to be symmetric, the direction of
the edge matters.
alg : string
Possible choices currently are:
* 'bp' for Loopy Belief Propagation.
* 'dd' for Dual Decomposition via Subgradients.
* 'trws' for Vladimirs TRWs implementation.
* 'trw' for OGM TRW.
* 'gibbs' for Gibbs sampling.
* 'lf' for Lazy Flipper
* 'fm' for Fusion Moves (alpha-expansion fusion)
* 'dyn' for Dynamic Programming (message passing in trees)
* 'gc' for Graph Cut
* 'alphaexp' for Alpha Expansion using Graph Cuts
* 'mqpbo' for multi-label qpbo
init : nd-array
Initial solution for starting inference (ignored by some algorithms).
reserveNumFactorsPerVariable :
reserve a certain number of factors for each variable can speed up
the building of a graphical model.
( For a 2d grid with second order factors one should set this to 5
4 2-factors and 1 unary factor for most pixels )
Returns
-------
labels : nd-array
Approximate (usually) MAP variable assignment.
"""
import opengm
n_states, pairwise_potentials = \
_validate_params(unary_potentials, pairwise_potentials, edges)
n_nodes = len(unary_potentials)
gm = opengm.gm(np.ones(n_nodes, dtype=opengm.label_type) * n_states)
nFactors = int(n_nodes + edges.shape[0])
gm.reserveFactors(nFactors)
gm.reserveFunctions(nFactors, 'explicit')
# all unaries as one numpy array
# (opengm's value_type == float64 but all types are accepted)
unaries = np.require(unary_potentials, dtype=opengm.value_type) * -1.0
# add all unart functions at once
fidUnaries = gm.addFunctions(unaries)
visUnaries = np.arange(n_nodes, dtype=opengm.label_type)
# add all unary factors at once
gm.addFactors(fidUnaries, visUnaries)
# add all pariwise functions at once
# - first axis of secondOrderFunctions iterates over the function)
secondOrderFunctions = -np.require(pairwise_potentials,
dtype=opengm.value_type)
fidSecondOrder = gm.addFunctions(secondOrderFunctions)
gm.addFactors(fidSecondOrder, edges.astype(np.uint64))
if alg == 'bp':
inference = opengm.inference.BeliefPropagation(gm)
elif alg == 'dd':
inference = opengm.inference.DualDecompositionSubgradient(gm)
elif alg == 'trws':
inference = opengm.inference.TrwsExternal(gm)
elif alg == 'trw':
inference = opengm.inference.TreeReweightedBp(gm)
elif alg == 'gibbs':
inference = opengm.inference.Gibbs(gm)
elif alg == 'lf':
inference = opengm.inference.LazyFlipper(gm)
elif alg == 'icm':
inference = opengm.inference.Icm(gm)
elif alg == 'dyn':
inference = opengm.inference.DynamicProgramming(gm)
elif alg == 'fm':
inference = opengm.inference.AlphaExpansionFusion(gm)
elif alg == 'gc':
inference = opengm.inference.GraphCut(gm)
elif alg == 'loc':
inference = opengm.inference.Loc(gm)
elif alg == 'mqpbo':
inference = opengm.inference.Mqpbo(gm)
elif alg == 'alphaexp':
inference = opengm.inference.AlphaExpansion(gm)
if init is not None:
inference.setStartingPoint(init)
inference.infer()
# we convert the result to int from unsigned int
# because otherwise we are sure to shoot ourself in the foot
res = inference.arg().astype(np.int)
if return_energy:
return res, gm.evaluate(res)
return res
import numpy
import opengm
import matplotlib.pyplot as plt
f1=numpy.ones([2])
f2=numpy.ones([2,2])
#Chain (non-shared functions):
numVar=5
gm=opengm.gm([2]*numVar)
for vi in xrange(numVar):
gm.addFactor(gm.addFunction(f1),vi)
if(vi+1<numVar):
gm.addFactor(gm.addFunction(f2),[vi,vi+1])
# visualize gm
opengm.visualizeGm( gm,show=False,layout='spring',plotFunctions=True,
plotNonShared=True,relNodeSize=0.4)
plt.savefig("chain_non_shared.png",bbox_inches='tight',dpi=300)
plt.close()
#Chain (shared high order functions):
numVar=5
gm=opengm.gm([2]*numVar)
fid2=gm.addFunction(f2)
for vi in xrange(numVar):
gm.addFactor(gm.addFunction(f1),vi)
if(vi+1<numVar):
gm.addFactor(fid2,[vi,vi+1])
# visualize gm
import matplotlib.pyplot as plt
f1=numpy.ones([2])
f2=numpy.ones([2,2])
"""
Grid:
- 4x4=16 variables
- second order factors in 4-neigbourhood
all connected to the same function
- higher order functions are shared
"""
size=3
gm=opengm.gm([2]*size*size)
fid=gm.addFunction(f2)
for y in range(size):
for x in range(size):
gm.addFactor(gm.addFunction(f1),x*size+y)
if(x+1<size):
gm.addFactor(fid,[x*size+y,(x+1)*size+y])
if(y+1<size):
gm.addFactor(fid,[x*size+y,x*size+(y+1)])
opengm.visualizeGm( gm,layout='spring',iterations=3000,
show=True,plotFunctions=True,
plotNonShared=True,relNodeSize=0.4)
plt.show
def inference_ogm(unary_potentials, pairwise_potentials, edges,
return_energy=False, alg='dd', init=None,
reserveNumFactorsPerVariable=2, **kwargs):
"""Inference with OpenGM backend.
Parameters
----------
unary_potentials : nd-array
Unary potentials of energy function.
pairwise_potentials : nd-array
Pairwise potentials of energy function.
edges : nd-array
Edges of energy function.
alg : string
Possible choices currently are:
* 'bp' for Loopy Belief Propagation.
* 'dd' for Dual Decomposition via Subgradients.
* 'trws' for Vladimirs TRWs implementation.
* 'trw' for OGM TRW.
* 'gibbs' for Gibbs sampling.
* 'lf' for Lazy Flipper
* 'fm' for Fusion Moves (alpha-expansion fusion)
* 'dyn' for Dynamic Programming (message passing in trees)
* 'gc' for Graph Cut
* 'alphaexp' for Alpha Expansion using Graph Cuts
* 'mqpbo' for multi-label qpbo
init : nd-array
Initial solution for starting inference (ignored by some algorithms).
reserveNumFactorsPerVariable :
reserve a certain number of factors for each variable can speed up
the building of a graphical model.
( For a 2d grid with second order factors one should set this to 5
4 2-factors and 1 unary factor for most pixels )
Returns
-------
labels : nd-array
Approximate (usually) MAP variable assignment.
"""
import opengm
n_states, pairwise_potentials = \
_validate_params(unary_potentials, pairwise_potentials, edges)
n_nodes = len(unary_potentials)
gm = opengm.gm(np.ones(n_nodes, dtype=opengm.label_type) * n_states)
nFactors = int(n_nodes + edges.shape[0])
gm.reserveFactors(nFactors)
gm.reserveFunctions(nFactors, 'explicit')
# all unaries as one numpy array
# (opengm's value_type == float64 but all types are accepted)
unaries = np.require(unary_potentials, dtype=opengm.value_type) * -1.0
# add all unart functions at once
fidUnaries = gm.addFunctions(unaries)
visUnaries = np.arange(n_nodes, dtype=opengm.label_type)
# add all unary factors at once
gm.addFactors(fidUnaries, visUnaries)
# add all pariwise functions at once
# - first axis of secondOrderFunctions iterates over the function)
secondOrderFunctions = -np.require(pairwise_potentials,
dtype=opengm.value_type)
fidSecondOrder = gm.addFunctions(secondOrderFunctions)
gm.addFactors(fidSecondOrder, edges.astype(np.uint64))
if alg == 'bp':
inference = opengm.inference.BeliefPropagation(gm)
elif alg == 'dd':
inference = opengm.inference.DualDecompositionSubgradient(gm)
elif alg == 'trws':
inference = opengm.inference.TrwsExternal(gm)
elif alg == 'trw':
inference = opengm.inference.TreeReweightedBp(gm)
elif alg == 'gibbs':
inference = opengm.inference.Gibbs(gm)
elif alg == 'lf':
inference = opengm.inference.LazyFlipper(gm)
elif alg == 'icm':
inference = opengm.inference.Icm(gm)
elif alg == 'dyn':
inference = opengm.inference.DynamicProgramming(gm)
elif alg == 'fm':
inference = opengm.inference.AlphaExpansionFusion(gm)
elif alg == 'gc':
inference = opengm.inference.GraphCut(gm)
elif alg == 'loc':
inference = opengm.inference.Loc(gm)
elif alg == 'mqpbo':
inference = opengm.inference.Mqpbo(gm)
elif alg == 'alphaexp':
inference = opengm.inference.AlphaExpansion(gm)
if init is not None:
inference.setStartingPoint(init)
inference.infer()
# we convert the result to int from unsigned int
# because otherwise we are sure to shoot ourself in the foot
res = inference.arg().astype(np.int)
if return_energy:
return res, gm.evaluate(res)
return res
tgrid = cgp2d.TopologicalGrid(seg.astype(numpy.uint64)) cgp = cgp2d.Cgp(tgrid) imgTopo = vigra.sampling.resize(imgLab,cgp.shape) imgRGBTopo = vigra.colors.transform_Lab2RGB(imgTopo) gradTopo = vigra.filters.gaussianGradientMagnitude(imgTopo,1.0) labelsTopo = vigra.sampling.resize(seg.astype(numpy.float32),cgp.shape,0) nVar = cgp.numCells(2) nFac = cgp.numCells(1) space = numpy.ones(nVar,dtype=opengm.label_type)*nVar gm = opengm.gm(space) wZero = numpy.zeros(nFac,dtype=opengm.value_type) pf = opengm.pottsFunctions([nVar,nVar],wZero,wZero) fids = gm.addFunctions(pf) gm.addFactors(fids,cgp.cell1BoundsArray()-1) cgc = opengm.inference.Cgc(gm=gm,parameter=opengm.InfParam(planar=True)) # visualize segmetation #cgp2d.visualize(img_rgb=imgRGBTopo,cgp=cgp)#,edge_data_in=bestState.astype(numpy.float32)) argDual = numpy.zeros(cgp.numCells(1),dtype=numpy.uint64)
import opengm import numpy chainLength=5 numLabels=100 numberOfStates=numpy.ones(chainLength,dtype=opengm.label_type)*numLabels gm=opengm.gm(numberOfStates,operator='adder') #add some random unaries for vi in range(chainLength): unaryFuction=numpy.random.random(numLabels) gm.addFactor(gm.addFunction(unaryFuction),[vi]) #add one 2.order function f=opengm.differenceFunction(shape=[numLabels]*2,weight=0.1) print type(f),f fid=gm.addFunction(f) #add factors on a chain for vi in range(chainLength-1): gm.addFactor(fid,[vi,vi+1]) inf = opengm.inference.BeliefPropagation(gm,parameter=opengm.InfParam(steps=40,convergenceBound=0 ,damping=0.9)) inf.infer(inf.verboseVisitor()) print inf.arg()
img = vigra.readImage(args[1]) if img.shape[2]!=3: print "Image must be RGB" sys.exit(0) T = float(args[6]) beta = float(args[7]) imgFlat = img.reshape([-1,3]).view(numpy.ndarray) numVar = imgFlat.shape[0] gm = opengm.gm(numpy.ones(numVar,dtype=opengm.label_type)*2) protoColor = numpy.array([args[3],args[4],args[5]],dtype=opengm.value_type).reshape([3,-1]) protoColor = numpy.repeat(protoColor,numVar,axis=1).swapaxes(0,1) diffArray = numpy.sum(numpy.abs(imgFlat - protoColor),axis=1) unaries = numpy.ones([numVar,2],dtype=opengm.value_type) unaries[:,0]=T unaries[:,1]=diffArray print diffArray gm.addFactors(gm.addFunctions(unaries),numpy.arange(numVar)) regularizer=opengm.pottsFunction([2,2],0.0,beta) gridVariableIndices=opengm.secondOrderGridVis(img.shape[0],img.shape[1])
