Mercurial > hg > nsaunier > traffic-intelligence

comparison scripts/train-object-classification.py @ 812:21f10332c72b

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
moved the classification parameters from tracking.cfg to a new classifier.cfg and made all classification parameters apparent
author Nicolas Saunier <nicolas.saunier@polymtl.ca>
date Fri, 10 Jun 2016 17:07:36 -0400
parents 52aa03260f03
children 85b81c46c526
comparison
equal deleted inserted replaced
810:082a5c2685f4 812:21f10332c72b
3 import numpy as np 3 import numpy as np
4 import sys, argparse 4 import sys, argparse
5 from cv2 import SVM_RBF, SVM_C_SVC 5 from cv2 import SVM_RBF, SVM_C_SVC
6 #from cv2.ml import SVM_RBF, SVM_C_SVC, ROW_SAMPLE # row_sample for layout in cv2.ml.SVM_load 6 #from cv2.ml import SVM_RBF, SVM_C_SVC, ROW_SAMPLE # row_sample for layout in cv2.ml.SVM_load
7 7
8 8 import cvutils, moving, ml, storage
9 import cvutils, moving, ml
10 9
11 parser = argparse.ArgumentParser(description='The program processes indicators for all pairs of road users in the scene') 10 parser = argparse.ArgumentParser(description='The program processes indicators for all pairs of road users in the scene')
12 parser.add_argument('-d', dest = 'directoryName', help = 'parent directory name for the directories containing the samples for the different road users', required = True) 11 parser.add_argument('-d', dest = 'directoryName', help = 'parent directory name for the directories containing the samples for the different road users', required = True)
13 parser.add_argument('--kernel', dest = 'kernelType', help = 'kernel type for the support vector machine (SVM)', default = SVM_RBF, type = long) 12 parser.add_argument('--kernel', dest = 'kernelType', help = 'kernel type for the support vector machine (SVM)', default = SVM_RBF, type = long)
14 parser.add_argument('--svm', dest = 'svmType', help = 'SVM type', default = SVM_C_SVC, type = long) 13 parser.add_argument('--svm', dest = 'svmType', help = 'SVM type', default = SVM_C_SVC, type = long)
15 # TODO make other SVM parameters apparent: C, C0, Nu, etc. 14 parser.add_argument('--deg', dest = 'degree', help = 'SVM degree', default = 0, type = int)
16 parser.add_argument('-s', dest = 'rescaleSize', help = 'rescale size of image samples', default = 64, type = int) 15 parser.add_argument('--gamma', dest = 'gamma', help = 'SVM gamma', default = 1, type = int)
17 parser.add_argument('-o', dest = 'nOrientations', help = 'number of orientations in HoG', default = 9, type = int) 16 parser.add_argument('--coef0', dest = 'coef0', help = 'SVM coef0', default = 0, type = int)
18 parser.add_argument('-p', dest = 'nPixelsPerCell', help = 'number of pixels per cell', default = 8, type = int) 17 parser.add_argument('--cvalue', dest = 'cvalue', help = 'SVM Cvalue', default = 1, type = int)
19 parser.add_argument('-c', dest = 'nCellsPerBlock', help = 'number of cells per block', default = 2, type = int) 18 parser.add_argument('--nu', dest = 'nu', help = 'SVM nu', default = 0, type = int)
19 parser.add_argument('--svmp', dest = 'svmP', help = 'SVM p', default = 0, type = int)
20 parser.add_argument('--cfg', dest = 'configFilename', help = 'name of the classifier configuration file', required = True)
21 # parser.add_argument('-s', dest = 'rescaleSize', help = 'rescale size of image samples', default = 64, type = int)
22 # parser.add_argument('-o', dest = 'nOrientations', help = 'number of orientations in HoG', default = 9, type = int)
23 # parser.add_argument('-p', dest = 'nPixelsPerCell', help = 'number of pixels per cell', default = 8, type = int)
24 # parser.add_argument('-c', dest = 'nCellsPerBlock', help = 'number of cells per block', default = 2, type = int)
25
20 args = parser.parse_args() 26 args = parser.parse_args()
27 classifierParams = storage.ClassifierParameters(args.configFilename)
21 28
22 rescaleSize = (args.rescaleSize, args.rescaleSize) 29 # rescaleSize = (args.rescaleSize, args.rescaleSize)
23 nPixelsPerCell = (args.nPixelsPerCell, args.nPixelsPerCell) 30 # nPixelsPerCell = (args.nPixelsPerCell, args.nPixelsPerCell)
24 nCellsPerBlock = (args.nCellsPerBlock, args.nCellsPerBlock) 31 # nCellsPerBlock = (args.nCellsPerBlock, args.nCellsPerBlock)
25 32
26 imageDirectories = {'pedestrian': args.directoryName + "/Pedestrians/", 33 imageDirectories = {'pedestrian': args.directoryName + "/Pedestrians/",
27 'bicycle': args.directoryName + "/Cyclists/", 34 'bicycle': args.directoryName + "/Cyclists/",
28 'car': args.directoryName + "/Vehicles/"} 35 'car': args.directoryName + "/Vehicles/"}
29 36
36 trainingSamplesPV = {} 43 trainingSamplesPV = {}
37 trainingLabelsPV = {} 44 trainingLabelsPV = {}
38 45
39 for k, v in imageDirectories.iteritems(): 46 for k, v in imageDirectories.iteritems():
40 print('Loading {} samples'.format(k)) 47 print('Loading {} samples'.format(k))
41 trainingSamples, trainingLabels = cvutils.createHOGTrainingSet(v, moving.userType2Num[k], rescaleSize, args.nOrientations, nPixelsPerCell, nCellsPerBlock) 48 trainingSamples, trainingLabels = cvutils.createHOGTrainingSet(v, moving.userType2Num[k], classifierParams.hogRescaleSize, classifierParams.hogNOrientations, classifierParams.hogNPixelsPerCell, classifierParams.hogNCellsPerBlock)
42 trainingSamplesPBV[k], trainingLabelsPBV[k] = trainingSamples, trainingLabels 49 trainingSamplesPBV[k], trainingLabelsPBV[k] = trainingSamples, trainingLabels
43 if k != 'pedestrian': 50 if k != 'pedestrian':
44 trainingSamplesBV[k], trainingLabelsBV[k] = trainingSamples, trainingLabels 51 trainingSamplesBV[k], trainingLabelsBV[k] = trainingSamples, trainingLabels
45 if k != 'car': 52 if k != 'car':
46 trainingSamplesPB[k], trainingLabelsPB[k] = trainingSamples, trainingLabels 53 trainingSamplesPB[k], trainingLabelsPB[k] = trainingSamples, trainingLabels
47 if k != 'bicycle': 54 if k != 'bicycle':
48 trainingSamplesPV[k], trainingLabelsPV[k] = trainingSamples, trainingLabels 55 trainingSamplesPV[k], trainingLabelsPV[k] = trainingSamples, trainingLabels
49 56
50 # Training the Support Vector Machine 57 # Training the Support Vector Machine
51 print "Training Pedestrian-Cyclist-Vehicle Model" 58 print "Training Pedestrian-Cyclist-Vehicle Model"
52 model = ml.SVM(args.svmType, args.kernelType) 59 model = ml.SVM(args.svmType, args.kernelType, args.degree, args.gamma, args.coef0, args.cvalue, args.nu, args.svmP)
53 model.train(np.concatenate(trainingSamplesPBV.values()), np.concatenate(trainingLabelsPBV.values())) 60 model.train(np.concatenate(trainingSamplesPBV.values()), np.concatenate(trainingLabelsPBV.values()))
54 model.save(args.directoryName + "/modelPBV.xml") 61 model.save(args.directoryName + "/modelPBV.xml")
55 62
56 print "Training Cyclist-Vehicle Model" 63 print "Training Cyclist-Vehicle Model"
57 model = ml.SVM(args.svmType, args.kernelType) 64 model = ml.SVM(args.svmType, args.kernelType, args.degree, args.gamma, args.coef0, args.cvalue, args.nu, args.svmP)
58 model.train(np.concatenate(trainingSamplesBV.values()), np.concatenate(trainingLabelsBV.values())) 65 model.train(np.concatenate(trainingSamplesBV.values()), np.concatenate(trainingLabelsBV.values()))
59 model.save(args.directoryName + "/modelBV.xml") 66 model.save(args.directoryName + "/modelBV.xml")
60 67
61 print "Training Pedestrian-Cyclist Model" 68 print "Training Pedestrian-Cyclist Model"
62 model = ml.SVM(args.svmType, args.kernelType) 69 model = ml.SVM(args.svmType, args.kernelType, args.degree, args.gamma, args.coef0, args.cvalue, args.nu, args.svmP)
63 model.train(np.concatenate(trainingSamplesPB.values()), np.concatenate(trainingLabelsPB.values())) 70 model.train(np.concatenate(trainingSamplesPB.values()), np.concatenate(trainingLabelsPB.values()))
64 model.save(args.directoryName + "/modelPB.xml") 71 model.save(args.directoryName + "/modelPB.xml")
65 72
66 print "Training Pedestrian-Vehicle Model" 73 print "Training Pedestrian-Vehicle Model"
67 model = ml.SVM(args.svmType, args.kernelType) 74 model = ml.SVM(args.svmType, args.kernelType, args.degree, args.gamma, args.coef0, args.cvalue, args.nu, args.svmP)
68 model.train(np.concatenate(trainingSamplesPV.values()), np.concatenate(trainingLabelsPV.values())) 75 model.train(np.concatenate(trainingSamplesPV.values()), np.concatenate(trainingLabelsPV.values()))
69 model.save(args.directoryName + "/modelPV.xml") 76 model.save(args.directoryName + "/modelPV.xml")