-
Notifications
You must be signed in to change notification settings - Fork 8
/
energy.py
210 lines (194 loc) · 9.57 KB
/
energy.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
# Image Completion using Statistics of Patch Offsets
# Author: Pranshu Gupta and Shrija Mishra
import cv2, numpy as np, sys, math, operator, maxflow, random, config as cfg
from scipy import ndimage
from time import time
from itertools import count, combinations
class Optimizer(object):
def __init__(self, image, mask, labels):
self.image = image/255.0
self.mask = mask
self.labels = labels
x, y = np.where(self.mask != 0)
sites = [[i, j] for (i, j) in zip(x, y)]
self.sites = sites
self.neighbors = []
self.dmem = np.zeros((len(sites), len(labels)))
self.InitializeD()
self.InitializeNeighbors()
def InitializeD(self):
for i in xrange(len(self.sites)):
for j in xrange(len(self.labels)):
self.dmem[i,j] = self.D(self.sites[i], self.labels[j])
def InitializeNeighbors(self):
start = time()
for i in xrange(len(self.sites)):
ne = []
neighbors = self.GetNeighbors(self.sites[i])
for n in neighbors:
if n in self.sites:
ne.append(self.sites.index(n))
self.neighbors.append(ne)
end = time()
print "InitializeNeighbors execution time: ", end - start
def D(self, site, offset):
i, j = site[0] + offset[0], site[1] + offset[1]
try:
if self.mask[i][j] == 0:
return 0
return float('inf')
except:
return float('inf')
def V(self, site1, site2, alpha, beta):
start = time()
x1a, y1a = site1[0] + alpha[0], site1[1] + alpha[1]
x2a, y2a = site2[0] + alpha[0], site2[1] + alpha[1]
x1b, y1b = site1[0] + beta[0], site1[1] + beta[1]
x2b, y2b = site2[0] + beta[0], site2[1] + beta[1]
try:
if self.mask[x1a, y1a] == 0 and self.mask[x1b, y1b] == 0 and self.mask[x2a, y2a] == 0 and self.mask[x2a, y2a] == 0:
return np.sum((self.image[x1a, y1a] - self.image[x1b, y1b])**2) + np.sum((self.image[x2a, y2a] - self.image[x2b, y2b])**2)
return 1000000.0
except:
return 1000000.0
def IsLowerEnergy(self, nodes, labelling1, labelling2):
updatedNodes = np.where(labelling1 != labelling2)[0]
diff = 0.0
for node in updatedNodes:
if self.D(self.sites[node], self.labels[labelling2[node]]) < float('inf'):
for n in self.neighbors[node]:
if n in updatedNodes:
if n > node:
diff += self.V(self.sites[node], self.sites[n], self.labels[labelling2[node]], self.labels[labelling2[n]]) - self.V(self.sites[node], self.sites[n], self.labels[labelling1[node]], self.labels[labelling1[n]])
else:
diff += self.V(self.sites[node], self.sites[n], self.labels[labelling2[node]], self.labels[labelling2[n]]) - self.V(self.sites[node], self.sites[n], self.labels[labelling1[node]], self.labels[labelling1[n]])
else:
return False
if diff < 0:
return True
return False
def GetNeighbors(self, site):
return [[site[0]-1, site[1]], [site[0], site[1]-1], [site[0]+1, site[1]], [site[0], site[1]+1]]
def AreNeighbors(self, site1, site2):
if np.abs(site1[0]-site2[0]) < 2 and np.abs(site1[1]-site2[1]) < 2:
return True
return False
def InitializeLabelling(self):
start = time()
labelling = [None]*len(self.sites)
for i in xrange(len(self.sites)):
perm = np.random.permutation(len(self.labels))
for j in perm:
if self.D(self.sites[i], self.labels[j]) < 1000000.0:
labelling[i] = j
break
self.sites = [self.sites[i] for i in range(len(self.sites)) if labelling[i] != None]
labelling = [label for label in labelling if label != None]
end = time()
print "InitializeLabelling execution time: ", end - start
return self.sites, np.array(labelling)
def CreateGraphABS(self, alpha, beta, ps, labelling):
start = time()
v = len(ps)
g = maxflow.Graph[float](v, 3*v)
nodes = g.add_nodes(v)
for i in range(v):
# add the data terms here
ta, tb = self.D(self.sites[ps[i]], self.labels[alpha]), self.D(self.sites[ps[i]], self.labels[beta])
# add the smoothing terms here
neighbor_list = self.neighbors[ps[i]]
for ind in neighbor_list:
try:
a, b, j = labelling[ps[i]], labelling[ind], ps.index(ind)
if j > i and (b == alpha or b == beta):
epq = self.V(self.sites[ps[i]], self.sites[ps[j]], self.labels[alpha], self.labels[beta])
g.add_edge(nodes[i], nodes[j], epq, epq)
else:
ea = self.V(self.sites[ps[i]], self.sites[ps[j]], self.labels[alpha], self.labels[b])
eb = self.V(self.sites[ps[i]], self.sites[ps[j]], self.labels[beta], self.labels[b])
ta, tb = ta + ea, tb + eb
except Exception as e:
pass
g.add_tedge(nodes[i], ta, tb)
end = time()
#print "CreateGraph execution time: ", end - start
return g, nodes
def CreateGraphAE(self, alpha, labelling):
start = time()
v = len(self.sites)
g = maxflow.Graph[float](2*v, 4*v)
nodes = g.add_nodes(v)
for i in range(v):
ta, tb = self.D(self.sites[i], self.labels[alpha]), float('inf')
if labelling[i] != alpha:
tb = self.D(self.sites[i], self.labels[labelling[i]])
g.add_tedge(nodes[i], ta, tb)
neighbor_list = self.neighbors[i]
for j in neighbor_list:
try:
if labelling[i] == labelling[j] and j > i:
epq = self.V(self.sites[i], self.sites[j], self.labels[labelling[i]], self.labels[alpha])
g.add_edge(nodes[i], nodes[j], epq, epq)
elif j > i:
aux_nodes = g.add_nodes(1)
epa = self.V(self.sites[i], self.sites[j], self.labels[labelling[i]], self.labels[alpha])
eaq = self.V(self.sites[i], self.sites[j], self.labels[labelling[j]], self.labels[alpha])
epq = self.V(self.sites[i], self.sites[j], self.labels[labelling[i]], self.labels[labelling[j]])
g.add_edge(nodes[i], aux_nodes[0], epa, epa)
g.add_edge(nodes[j], aux_nodes[0], eaq, eaq)
g.add_tedge(aux_nodes[0], float('inf'), epq)
except Exception as e:
print(e)
end = time()
#print "CreateGraph execution time: ", end - start
return g, nodes
def OptimizeLabellingABS(self, labelling):
labellings = np.zeros((2, len(self.sites)), dtype=int)
labellings[0] = labellings[1] = np.copy(labelling)
iter_count = 0
while(True):
start = time()
success = 0
for alpha, beta in combinations(range(len(self.labels)), 2):
ps = [i for i in range(len(self.sites)) if (labellings[0][i] == alpha or labellings[0][i] == beta)]
if len(ps) > 0:
g, nodes = self.CreateGraphABS(alpha, beta, ps, labellings[0])
flow = g.maxflow()
for i in range(len(ps)):
gamma = g.get_segment(nodes[i])
labellings[1, ps[i]] = alpha*(1-gamma) + beta*gamma
if self.IsLowerEnergy(ps, labellings[0], labellings[1]):
labellings[0, ps] = labellings[1, ps]
success = 1
else:
labellings[1, ps] = labellings[0, ps]
iter_count += 1
end = time()
print "ABS Iteration " + str(iter_count) + " execution time: ", str(end - start)
if success != 1 or iter_count >= cfg.MAX_ITER:
break
return labellings[0]
def OptimizeLabellingAE(self, labelling):
labellings = np.zeros((2, len(self.sites)), dtype=int)
labellings[0] = labellings[1] = np.copy(labelling)
iter_count = 0
while(True):
start = time()
success = 0
for alpha in xrange(len(self.labels)):
g, nodes = self.CreateGraphAE(alpha, labellings[0])
flow = g.maxflow()
for i in range(len(self.sites)):
gamma = g.get_segment(nodes[i])
labellings[1, i] = alpha*(1-gamma) + labellings[1, i]*gamma
if self.IsLowerEnergy(range(len(self.sites)), labellings[0], labellings[1]):
labellings[0] = labellings[1]
success = 1
else:
labellings[1] = labellings[0]
iter_count += 1
end = time()
print "AE Iteration " + str(iter_count) + " execution time: ", str(end - start)
if success != 1 or iter_count >= cfg.MAX_ITER:
break
return labellings[0]