-
Notifications
You must be signed in to change notification settings - Fork 0
/
supra.py
392 lines (312 loc) · 11.2 KB
/
supra.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
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
import numpy as np, pickle, random, pxutil, copy, math, sqlitedict, os, tempfile
import normalisedImage, normalisedImageOpt, simpleGbrt, supraFeatures
import skimage.color as col, skimage.feature as feature, skimage.filter as filt
from sklearn.ensemble import GradientBoostingRegressor
import matplotlib.pyplot as plt
def SignAgreement(testOff, testPred):
signTotal = 0
for tru, pred in zip(testOff, testPred):
truSign = tru >= 0.
predSign = pred >= 0.
if truSign == predSign:
signTotal += 1
signScore = float(signTotal) / len(testOff)
return signScore
class SupraAxis():
def __init__(self, axisXIn = 1., axisYIn = 0.):
self.reg = None
self.x = axisXIn
self.y = axisYIn
def PrepareModel(self, features, offsets):
if self.reg is not None:
return 0
self.reg = GradientBoostingRegressor()
offsets = np.array(offsets)
labels = offsets[:,0] * self.x + offsets[:,1] * self.y
if not np.all(np.isfinite(labels)):
raise Exception("Training labels contains non-finite value(s), either NaN or infinite")
self.reg.fit(features, labels)
def IsModelReady(self):
return self.reg is not None
def ClearModel(self):
self.reg = None
def GetFeatureImportance(self):
return self.reg.feature_importances_
def ListCompare(la, lb):
if la is None and lb is None: return True
if la is None: return False
if lb is None: return False
if len(la) != len(lb): return False
for a, b in zip(la, lb):
if a != b: return False
return True
class SupraAxisSet():
def __init__(self, ptNumIn, numPoints = 5, supportPixHalfWidthIn = 0.3, numSupportPix = 50):
self.ptNum = ptNumIn
self.supportPixOff = None
self.supportPixOffSobel = None
self.trainInt = []
self.trainOffX, self.trainOffY = [], []
self.regX, self.regY = None, None
self.sobelKernel = np.array([[1,0,-1],[2,0,-2],[1,0,-1]], dtype=np.int32)
#self.sobelOffsets, self.sobelCoeffs, halfWidth = normalisedImageOpt.CalcKernelOffsets(self.sobelKernel)
#self.featureMultiplex = simpleGbrt.FeatureGenTest()
self.trainIntDb = None
self.numSupportPix = numSupportPix
self.featureGen = supraFeatures.FeatureGen(numPoints, supportPixHalfWidthIn, numSupportPix, 1)
self.numPoints = numPoints
self.featureMask = None
self.axes = None
def __del__(self):
del self.trainIntDb
try:
if self.trainIntDbFina is not None:
os.remove(self.trainIntDbFina)
except:
pass
def SetParameters(self, params):
pass
def IsModelReady(self):
if self.axes is None:
return False
countUnready = 0
for axis in self.axes:
if not axis.IsModelReady():
countUnready += 1
if countUnready > 0: return False
return True
def AddTraining(self, sample, trainOffset, extraFeatures):
#Check at least one axis requires data
if self.IsModelReady(): return 0
xOff = trainOffset[self.ptNum][0]
yOff = trainOffset[self.ptNum][1]
if self.trainIntDb is None:
self.trainIntDbFina = tempfile.mkstemp()[1]
self.trainIntDb = sqlitedict.SqliteDict(self.trainIntDbFina, autocommit=True)
self.featureGen.SetImage(sample)
self.featureGen.SetModel(np.array(sample.procShape))
self.featureGen.SetModelOffset(trainOffset)
self.featureGen.SetShapeNoise(0.6)
self.featureGen.SetPointNum(self.ptNum)
self.featureGen.SetOffset(xOff, yOff)
self.featureGen.Gen()
feat = self.featureGen.GetGenFeat()
if feat.shape[0] != len(self.featureMask):
print "Warning: Generated features have incorrect number of components, got "+\
str(feat.shape[0])+", received "+str(len(self.featureMask))
featComp = np.concatenate((feat, extraFeatures))
if not np.all(np.isfinite(featComp)):
for i, comp in enumerate(featComp):
print comp, math.isinf(comp), math.isnan(comp),
if i < len(self.featureMask):
print self.featureMask[i]
else:
print ""
raise Exception("Training data contains non-finite value(s), either NaN or infinite (1)")
#self.trainInt.append(features)
self.trainIntDb[str(len(self.trainOffX))] = featComp
self.trainOffX.append(xOff)
self.trainOffY.append(yOff)
return 1
def PrepareModel(self):
#Check at least one axis requires data
if self.IsModelReady(): return 0
self.axes = []
self.axes.append(SupraAxis(1., 0.))
self.axes.append(SupraAxis(0., 1.))
trainOff = np.vstack([self.trainOffX, self.trainOffY]).transpose()
keys = map(int, self.trainIntDb.keys())
print "Loading",len(keys),"samples for training"
keys.sort()
self.trainInt = np.empty((len(keys), len(self.trainIntDb[0])), dtype=np.float32, order='C')
for k in keys:
self.trainInt[k, :] = self.trainIntDb[str(k)]
if not np.all(np.isfinite(self.trainInt)):
raise Exception("Training data contains non-finite value(s), either NaN or infinite (2)")
for axis in self.axes:
axis.PrepareModel(self.trainInt, trainOff)
self.trainInt = None
return 1
def ClearTraining(self):
self.trainInt = None
self.trainIntDb = None
try:
os.remove(self.trainIntDbFina)
except:
pass
self.trainOffX, self.trainOffY = [], []
def GetFeatureImportance(self):
out = []
for axis in self.axes:
out.append(axis.GetFeatureImportance())
return out
def Predict(self, sample, model, prevFrameFeatures):
self.featureGen.SetImage(sample)
self.featureGen.SetModel(np.array(model))
self.featureGen.ClearModelOffset()
self.featureGen.SetShapeNoise(0.)
self.featureGen.SetPointNum(self.ptNum)
self.featureGen.SetOffset(0., 0.)
self.featureGen.Gen()
feat = self.featureGen.GetGenFeat()
featComp = np.concatenate((feat, prevFrameFeatures))
#self.featureMultiplex.ClearFeatureSets()
#self.featureMultiplex.AddFeatureSet(self.featureGen.GetGenFeat())
totalx, totaly, weightx, weighty = 0., 0., 0., 0.
for axis in self.axes:
#pred = simpleGbrt.PredictGbrt(axis.reg, self.featureMultiplex)
pred = axis.reg.predict([featComp])[0]
totalx += pred * axis.x
totaly += pred * axis.y
weightx += axis.x
weighty += axis.y
return totalx / weightx, totaly / weighty
def SetFeatureMask(self, mask):
#if not ListCompare(mask, self.featureMask):#Hack, this should be done selectively
if 1:
self.featureMask = mask
self.featureGen.SetFeatureMask(mask)
self.ClearModels()
return 1
return 0
def GetFeatureList(self):
return self.featureGen.GetFeatureList()
def ClearModels(self):
if self.axes is not None:
for axis in self.axes:
axis.ClearModel()
class SupraCloud():
def __init__(self, supportPixHalfWidthIn = 0.3, trainingOffsetIn = 0.3, numPoints = 5):
self.trainingOffset = trainingOffsetIn #Standard deviations
self.supportPixHalfWidth = supportPixHalfWidthIn
self.numIter = 2
self.numPoints = numPoints
self.trackers = []
self.featureGen = []
for i in range(self.numPoints):
self.trackers.append(SupraAxisSet(i, self.numPoints, self.supportPixHalfWidth))
def SetParameters(self, params):
if 'trainingOffset' in params:
#if self.trainingOffset != params['trainingOffset']: #Hack to check effect
# self.ClearModels()
self.trainingOffset = params['trainingOffset']
print "trainingOffset=", self.trainingOffset
def AddTraining(self, sample, numExamples, extraFeatures):
mod = 0
for sampleCount in range(numExamples):
perturb = []
for num in range(sample.NumPoints()):
perturb.append((np.random.normal(scale=self.trainingOffset),\
np.random.normal(scale=self.trainingOffset)))
for count, tracker in enumerate(self.trackers):
mod += tracker.AddTraining(sample, perturb, extraFeatures)
if mod == 0:
raise Exception("No training added")
def PrepareModel(self):
mod = 0
for tracker in self.trackers:
mod += tracker.PrepareModel()
if mod == 0:
raise Exception("No change in model")
def ClearTraining(self):
for tracker in self.trackers:
tracker.ClearTraining()
def IsModelReady(self):
for tracker in self.trackers:
if not tracker.IsModelReady(): return False
return True
def GetFeatureImportance(self):
out = []
for tracker in self.trackers:
out.extend(tracker.GetFeatureImportance())
return out
def Predict(self, sample, model, prevFrameFeatures):
currentModel = np.array(copy.deepcopy(model))
for iterNum in range(self.numIter):
for ptNum, tracker in enumerate(self.trackers):
pred = tracker.Predict(sample, currentModel, prevFrameFeatures)
currentModel[ptNum,:] -= pred
return currentModel
def SetFeatureMasks(self, masks):
changed = 0
if len(self.trackers) != len(masks):
raise Exception("Number of trackers is incorrect")
for tracker, mask in zip(self.trackers, masks):
changed += tracker.SetFeatureMask(mask)
return changed
def GetFeatureList(self):
masks = []
for tracker in self.trackers:
masks.append(tracker.GetFeatureList())
return masks
def ClearModels(self):
for tracker in self.trackers:
tracker.ClearModels()
class SupraLayers:
def __init__(self, trainNormSamples):
self.numPoints = trainNormSamples[0].NumPoints()
self.featureGenPrevFrame = supraFeatures.FeatureGenPrevFrame(trainNormSamples, 20, self.numPoints)
if self.numPoints == 0:
raise ValueError("Model must have non-zero number of points")
for sample in trainNormSamples:
if sample.NumPoints() != self.numPoints:
raise ValueError("Model must have consistent number of points")
self.layers = [SupraCloud(0.3,0.2,self.numPoints),SupraCloud(0.3,0.05,self.numPoints)]
def SetParameters(self, params):
if params is None:
raise Exception("Invalid input")
if len(params) != len(self.layers):
raise Exception("Incorrect number of param layers")
for player, layer in zip(params, self.layers):
layer.SetParameters(player)
def AddTraining(self, sample, numExamples):
#Add noise to shape for previous frame features
prevShapePerturb = copy.deepcopy(sample.procShape)
for ptNum in range(len(prevShapePerturb)):
pos = prevShapePerturb[ptNum]
pos[0] += np.random.normal(scale=0.1)
pos[1] += np.random.normal(scale=0.1)
#Extract features from synthetic previous frame
extraFeatures = self.featureGenPrevFrame.Gen(sample, prevShapePerturb)
for layer in self.layers:
layer.AddTraining(sample, numExamples, extraFeatures)
def PrepareModel(self):
for layer in self.layers:
layer.PrepareModel()
def ClearTraining(self):
for layer in self.layers:
layer.ClearTraining()
def IsModelReady(self):
for layer in self.layers:
if not layer.IsModelReady(): return False
return True
def GetFeatureImportance(self):
out = []
for layer in self.layers:
out.extend(layer.GetFeatureImportance())
return out
def CalcPrevFrameFeatures(self, sample, model):
#Extract features from synthetic previous frame
return self.featureGenPrevFrame.Gen(sample, model)
def Predict(self, sample, model, prevFrameFeatures):
currentModel = np.array(copy.deepcopy(model))
for layerNum, layer in enumerate(self.layers):
currentModel = layer.Predict(sample, currentModel, prevFrameFeatures)
return currentModel
def SetFeatureMasks(self, masks):
changed = 0
if len(self.layers) != len(masks):
raise Exception("Number of layers is incorrect")
for layer, masksIt in zip(self.layers, masks):
changed += layer.SetFeatureMasks(masksIt)
return changed
def GetFeatureList(self):
out = []
for layer in self.layers:
out.append(layer.GetFeatureList())
return out
def ClearModels(self):
for layer in self.layers:
layer.ClearModels()
if __name__ == "__main__":
pass