def MRF_modelling(superpixels, labels, unaries, data, p=0.2):
edges, surfaces = adjGraph(superpixels, data, False)
# p : smoothing parameter
# Labels : is supposed to be an array: length = #SP, width= #Labels with the probaility for each label and each SP
nLabels = labels.shape[1] # The width of the labels array is the number of different labels
nVariables = superpixels.max() # The max number of the sp ID is the total number of SPs
nEdges = edges.shape[0] # Number of edges = lenght of edges-array
edgeVis = edges[:,0:2] # Vertice-Tuples in Edges list
unaryValues= labels
sameLabel = np.zeros((nVariables,1))
diffLabel = np.ones((nVariables,1)) * p
edgeValues = np.concatenate((sameLabel, diffLabel), axis=1)
gm = getGraphicalModel(
nLabels = nLabels,
nVariables = nVariables,
nEdges = nEdges,
edgeVis = edgeVis,
unaryValues = unaryValues,
edgeValues = edgeValues,
gmOperator = 'adder')
gm
return gm
def getData():
l = h5py.File("data/labels.h5", "r")
labels = l["labels"].value
labels = labels/labels.sum(axis=1)
f = h5py.File('data/block00.h5', 'r')
data = f['volume/data'].value
d = h5py.File("data/ws.h5",'r')
sp = d['ws'].value
edges = adjGraph(sp, data, False)
return data, sp, edges, labels
def testAdjGraph():
f = h5py.File("data/block00.h5", 'r') # load image data
d = f["volume/data"].value
ws = h5py.File("data/ws.h5", 'r') # load watershed segmentation / superpixels
sp = ws["ws"].value
edges, surfaces = adjGraph(sp, d.astype(np.float32), True) # compute region adjecency graph
g = h5py.File("data/edges.h5", 'w')
g.create_dataset("edges", data=edges)
g.close()
# e = h5py.File("data/edges.h5", 'r')
# edges = e["edges"].value
pickle.dump(surfaces, open("data/surfaces.p", "wb"))
def MRF_modelling(superpixels, labels, unaries, data):
edges, surfaces = adjGraph(superpixels, data)
nLabels = labels.shape[0]#################################################
nVariables = superpixels.max()###############################################
nEdges = edges.shape[0]
edgeVis = edges[:,0:2]
unaryValues= unaries#########################################################
edgeValues = binaryClassifier(edges)#########################################
gm = getGraphicalModel(
nLabels = nLabels,
nVariables = nVariables,
nEdges = nEdges,
edgeVis = edgeVis,
unaryValues = unaryValues,
edgeValues = edgeValues,
gmOperator = 'adder')
gm
return gm
# -*- coding: utf-8 -*-
"""
Created on Sat Jun 29 14:00:29 2013
@author: mfk
"""
import numpy as np
import _adjGraph as gm
from mpl_toolkits.mplot3d import Axes3D
from matplotlib import pyplot as plt
test1 = np.zeros((3,3,3),dtype=np.uint32)
test1[:,:,0] = 1
result = gm.adjGraph(test1, test1.astype(np.float32), False)
points = result[1][0]
fig = plt.figure("EdgePoints", figsize=(10,10)) # create a figure
ax = fig.add_subplot(111, projection="3d") # create subplot 3d in figure
ax.scatter(points[:,0], points[:,1], points[:,2])
print result
# Test functions for the adjGraph-search algorithm #
test1 = np.zeros((3,3,3),dtype=np.uint32)
result = gm.adjGraph(test1, test1.astype(np.float32), True) # should return empty edges and #edges=0
print result
test1[:,:,0] = 1
result = gm.adjGraph(test1, test1.astype(np.float32), True) # should return edge (0,1) and #edges=1
print result