-
Notifications
You must be signed in to change notification settings - Fork 4
/
gmm_fit3.py
249 lines (223 loc) · 8.86 KB
/
gmm_fit3.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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
import numpy as np
from scipy.stats import multivariate_normal as mvn_pdf
import matplotlib.pyplot as plt
from cluster import MiniBatchKMeans
from mixture import GaussianMixture
import pymesh
from scipy.special import logsumexp
mesh0 = pymesh.load_mesh("bunny/bun_zipper_res4.ply")
#mesh3 = pymesh.load_mesh("bunny/bun_zipper_res4_pds.ply")
#mesh4 = pymesh.load_mesh("bunny/bun_zipper_res4_25k_pds.ply")
mesh4 = pymesh.load_mesh("bunny/bun_zipper_res4_sds.ply")
def get_centroids(mesh):
# obtain a vertex for each face index
face_vert = mesh.vertices[mesh.faces.reshape(-1),:].reshape((mesh.faces.shape[0],3,-1))
# face_vert is size (faces,3(one for each vert), 3(one for each dimension))
centroids = face_vert.sum(1)/3.0
ABAC = face_vert[:,1:3,:] - face_vert[:,0:1,:]
areas = np.linalg.norm(np.cross(ABAC[:,0,:],ABAC[:,1,:]),axis=1)/2.0
return centroids, areas
com,a = get_centroids(mesh0)
face_vert = mesh0.vertices[mesh0.faces.reshape(-1),:].reshape((mesh0.faces.shape[0],3,-1))
#gm3 = GaussianMixture(100,init_params='kmeans'); gm3.set_triangles(face_vert); gm3.fit(com); gm3.set_triangles(None)
gm3 = GaussianMixture(1,init_params='kmeans',tol=1e-4,max_iter=100); gm3.fit(mesh4.vertices)
def tri_loss(gmm,faces_and_verts):
centroids = face_vert.mean(1)
ABAC = face_vert[:,1:3,:] - face_vert[:,0:1,:]
areas = np.linalg.norm(np.cross(ABAC[:,0,:],ABAC[:,1,:]),axis=1)/2.0
#areas = areas/areas.sum()
total = 0.0
#for idx, face in enumerate(faces_and_verts):
#face is 3 faces with 3d locs
#center = face.mean(0)
#centr2 = centroids[idx,:]
A = faces_and_verts[:,0,:]
B = faces_and_verts[:,1,:]
C = faces_and_verts[:,2,:]
#m = center.reshape((-1,1))
#thing = np.zeros(gmm.weights_.shape)
thing = np.zeros((faces_and_verts.shape[0],gmm.weights_.shape[0]))
i = 0
#things =
weights = np.zeros(thing.shape)
for mu, s, si, pi in zip(gmm.means_,gmm.covariances_,gmm.precisions_,gmm.weights_):
weights[:,i] = mvn_pdf.pdf(centroids,mu,s)
#print(mvn_pdf.pdf(points,mu,s).shape,weights.shape)
i+=1
row_sums = weights.sum(axis=1)
#print(row_sums.shape)
weights = weights / row_sums[:, np.newaxis]
i=0
for mu, s, si, pi in zip(gmm.means_,gmm.covariances_,gmm.precisions_,gmm.weights_):
res = 0.0
dev = (centroids - mu)
res = 0.0
res -= 0.5 * np.log(2*np.pi) *3
res -= 0.5 * np.log(np.linalg.det(s))
t1 = (dev.dot(si)*dev).sum(1)
t2 = (A.dot(si)*A + B.dot(si)*B + C.dot(si)*C - 3*centroids.dot(si)*centroids).sum(1)
#print("T1\t",t1.sum(),t1.min(),t1.max(),t1.mean())
#print("T2\t",t2.sum(),t2.min(),t2.max(),t2.mean())
res -= 0.5 * (t1 + (1.0/12.0) * t2)
total += ((res + np.log(pi))).sum()
thing[:,i] = ((res+ np.log(pi)))
i+=1
#total += thing.sum()*#areas[idx]#logsumexp(thing)*areas[idx]
return logsumexp(thing,axis=1).mean()#.sum()/areas.sum()#.mean()#/points.shape[0]
#return total/areas.sum()#faces_and_verts.shape[0]
def tri_loss_lb(gmm,faces_and_verts):
centroids = face_vert.mean(1)
ABAC = face_vert[:,1:3,:] - face_vert[:,0:1,:]
areas = np.linalg.norm(np.cross(ABAC[:,0,:],ABAC[:,1,:]),axis=1)/2.0
#areas = areas/areas.sum()
total = 0.0
#for idx, face in enumerate(faces_and_verts):
#face is 3 faces with 3d locs
#center = face.mean(0)
#centr2 = centroids[idx,:]
A = faces_and_verts[:,0,:]
B = faces_and_verts[:,1,:]
C = faces_and_verts[:,2,:]
#m = center.reshape((-1,1))
#thing = np.zeros(gmm.weights_.shape)
thing = np.zeros((faces_and_verts.shape[0],gmm.weights_.shape[0]))
i = 0
#things =
weights = np.zeros(thing.shape)
for mu, s, si, pi in zip(gmm.means_,gmm.covariances_,gmm.precisions_,gmm.weights_):
weights[:,i] = mvn_pdf.pdf(centroids,mu,s)
#print(mvn_pdf.pdf(points,mu,s).shape,weights.shape)
i+=1
row_sums = weights.sum(axis=1)
#print(row_sums.shape)
weights = weights / row_sums[:, np.newaxis]
i=0
for mu, s, si, pi in zip(gmm.means_,gmm.covariances_,gmm.precisions_,gmm.weights_):
res = 0.0
dev = (centroids - mu)
res = 0.0
res -= 0.5 * np.log(2*np.pi) *3
res -= 0.5 * np.log(np.linalg.det(s))
t1 = (dev.dot(si)*dev).sum(1)
t2 = (A.dot(si)*A + B.dot(si)*B + C.dot(si)*C - 3*centroids.dot(si)*centroids).sum(1)
#print("T1\t",t1.sum(),t1.min(),t1.max(),t1.mean())
#print("T2\t",t2.sum(),t2.min(),t2.max(),t2.mean())
res -= 0.5 * (t1 + (1.0/12.0) * t2)
total += ((res + np.log(pi))).sum()
thing[:,i] = ((res+ np.log(pi)))*areas#/areas.mean()
i+=1
#total += thing.sum()*#areas[idx]#logsumexp(thing)*areas[idx]
#thing = thing*weights
return np.sum(thing,axis=1).sum()/areas.sum()#.sum()/areas.sum()#.mean()#/points.shape[0]
#return total/areas.sum()#faces_and_verts.shape[0]
def pt_loss(gmm,points):
total = 0.0
#for p in points:
thing = np.zeros((points.shape[0],gmm.weights_.shape[0]))
i = 0
#things =
for mu, s, si, pi in zip(gmm.means_,gmm.covariances_,gmm.precisions_,gmm.weights_):
res = 0.0
dev = points-mu
res = 0.0
res -= 0.5 * np.log(2*np.pi) *3
res -= 0.5 * np.log(np.linalg.det(s))
t1 = (dev.dot(si) * dev).sum(1)
res -= 0.5 * t1
#total += (res + np.log(pi)).sum()
thing[:,i] = (res + np.log(pi))
i+=1
#total += thing.sum()#logsumexp(thing)
return logsumexp(thing,axis=1).mean()#logsumexp(thing,axis=1).mean()#/points.shape[0]
def pt_loss_lb(gmm,points):
total = 0.0
#for p in points:
thing = np.zeros((points.shape[0],gmm.weights_.shape[0]))
i = 0
#things =
weights = np.zeros(thing.shape)
for mu, s, si, pi in zip(gmm.means_,gmm.covariances_,gmm.precisions_,gmm.weights_):
weights[:,i] = mvn_pdf.pdf(points,mu,s)
#print(mvn_pdf.pdf(points,mu,s).shape,weights.shape)
i+=1
row_sums = weights.sum(axis=1)
#print(row_sums.shape)
weights = weights / row_sums[:, np.newaxis]
i=0
for mu, s, si, pi in zip(gmm.means_,gmm.covariances_,gmm.precisions_,gmm.weights_):
res = 0.0
dev = points-mu
res = 0.0
res -= 0.5 * np.log(2*np.pi) *3
res -= 0.5 * np.log(np.linalg.det(s))
t1 = (dev.dot(si) * dev).sum(1)
res -= 0.5 * t1
#total += (res + np.log(pi)).sum()
thing[:,i] = (res + np.log(pi))
i+=1
#total += thing.sum()#logsumexp(thing)
#thing = thing*weights
return thing.sum(axis=1).mean()#logsumexp(thing,axis=1).mean()#/points.shape[0]
def com_loss(gmm,points,areas):
total = 0.0
#for p in points:
thing = np.zeros((points.shape[0],gmm.weights_.shape[0]))
i = 0
#things =
for mu, s, si, pi in zip(gmm.means_,gmm.covariances_,gmm.precisions_,gmm.weights_):
res = 0.0
dev = points-mu
res = 0.0
res -= 0.5 * np.log(2*np.pi) *3
res -= 0.5 * np.log(np.linalg.det(s))
t1 = (dev.dot(si) * dev).sum(1)
res -= 0.5 * t1
#total += (res + np.log(pi)).sum()
thing[:,i] = (res + np.log(pi))*(areas/areas.mean())
i+=1
#total += thing.sum()#logsumexp(thing)
return logsumexp(thing,axis=1).mean()#/points.shape[0]
def com_loss_lb(gmm,points,areas):
total = 0.0
#for p in points:
thing = np.zeros((points.shape[0],gmm.weights_.shape[0]))
i = 0
#things =
for mu, s, si, pi in zip(gmm.means_,gmm.covariances_,gmm.precisions_,gmm.weights_):
res = 0.0
dev = points-mu
res = 0.0
res -= 0.5 * np.log(2*np.pi) *3
res -= 0.5 * np.log(np.linalg.det(s))
t1 = (dev.dot(si) * dev).sum(1)
res -= 0.5 * t1
#total += (res + np.log(pi)).sum()
thing[:,i] = (res + np.log(pi))*(areas/areas.mean())
i+=1
#total += thing.sum()#logsumexp(thing)
return np.sum(thing,axis=1).mean()#/points.shape[0]
if True:
tl = tri_loss_lb
cl = com_loss_lb
pl = pt_loss_lb
print("OMG")
else:
tl = tri_loss
cl = com_loss
pl = pt_loss
print("tri\t",tl(gm3,face_vert),'\t',0)
print("mpt\t",pl(gm3,com),'\t',0)
print('com\t',cl(gm3,com,a),'\t',0)
#print("ptLB\t",pt_loss_lb(gm3,com))
#print("spt\t",gm3.score(com))
#print("sp\t",gm3._estimate_weighted_log_prob(com).sum())
for pn in np.logspace(1,np.log10(mesh4.vertices.shape[0]*.95),10):
scores = []
for itern in range(10):
ptsn = np.random.choice(range(mesh4.vertices.shape[0]),int(pn),replace=False)
scores.append(pl(gm3,mesh4.vertices[ptsn,:]))
#scores.append(gm3._estimate_weighted_log_prob(mesh4.vertices[ptsn,:]).sum()/pn)
scores = np.array(scores)
print(ptsn.shape[0],'\t',scores.mean(),'\t',scores.std())
#print(" ",gm3.score(mesh4.vertices))
#print(" ",gm3._estimate_weighted_log_prob(mesh4.vertices).sum()/mesh4.vertices.shape[0])