def create_params_from_psi(w):
# If this changes then we also need to update
# - init_constraints() so that the right weights are positive
# - TrainingInstance::GetFeatureForHypothesis()
wts = np.array(w[:-2], float)
# The factor of 100 here _must_ match that in ComputeFeatureForHypothesis()
return py_indoor_context.ManhattanHyperParameters(wts,
w[-2]*kPenaltyFactor,
w[-1]*kPenaltyFactor)
class ICCV:
StereoOffsets = ([-1, 1])
NumAuxFrames = len(StereoOffsets)
StereoWeights = np.array([1. / NumAuxFrames] * NumAuxFrames)
Weights = np.hstack([.001, 100., 5., 50. * StereoWeights])
CornerPenalty = 70.
OcclusionPenalty = 50.
Params = py_indoor_context.ManhattanHyperParameters(
Weights, CornerPenalty, OcclusionPenalty)
import py_indoor_context
from matplotlib.pyplot import *
import training_helpers
np.set_printoptions(precision=3)
w = np.array([
1.,
-1.,
0.,
])
params = py_indoor_context.ManhattanHyperParameters(w, 4., 6.)
w2 = np.array([
2.,
-.5,
.1,
])
params2 = py_indoor_context.ManhattanHyperParameters(w2, 3., 7.)
train_ids = [20, 40, 60]
test_ids = [65, 70]
mgr = py_indoor_context.TrainingManager()
mgr.LoadSequence("lab_kitchen1", train_ids + test_ids)
fm = py_indoor_context.FeatureManager('/tmp')
for i in range(mgr.NumInstances()):
fm.ComputeMockFeatures(mgr.GetInstance(i))
import numpy as np
from matplotlib.pyplot import *
import py_indoor_context
import training_helpers
np.set_printoptions(precision=3)
#w = np.array([0., 0., 1.,])
w = np.array([0.331, -0.662, 0.004])
#w = np.array([229.582, 147.414, -11.827, 75.342, 87.373, 27.623, 37.698])
#params = py_indoor_context.ManhattanHyperParameters(w, 400., 600.)
params = py_indoor_context.ManhattanHyperParameters(w, 13.462474823, 0.)
mgr = py_indoor_context.TrainingManager()
#mgr.LoadSequence('lab_kitchen1', [25]) #range(10,80,10))
mgr.LoadSequence('exeter_mcr1', [35]) #range(10,80,10))
#mgr.ComputeAllFeatures([-5,-1,1,5], True)
mgr.ComputeMockFeatures()
instances = [ mgr.GetInstance(i) for i in range(mgr.GetNumInstances()) ]
sum_err = 0.
for inst in instances:
inst.ConfigureLabellingLoss()
print '\nFrame ',inst.GetFrameId()
payoffs = py_indoor_context.DPPayoffs()
inst.CompileFeatures(params, payoffs)
hyp = mgr.Solve(inst, payoffs)
class ECCV2010_v2:
Weights = np.array([1.])
CornerPenalty = 10.
OcclusionPenalty = 0.
Params = py_indoor_context.ManhattanHyperParameters(
Weights, CornerPenalty, OcclusionPenalty)