Mercurial Hosting > traffic-intelligence

comparison scripts/train-object-classification.py @ 519:4ad5123d969e

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
added script to train HoG-SVM classifiers for object classification (based on a script by Sohail Zangenehpour, PhD student at McGill)
author Nicolas Saunier <nicolas.saunier@polymtl.ca>
date Wed, 18 Jun 2014 00:53:39 -0400
parents
children ce40a89bd6ae
comparison
equal deleted inserted replaced
518:0c86c73f3c09 519:4ad5123d969e
1 #! /usr/bin/env python
2
3 import numpy as np
4 import sys, argparse
5 from cv2 import SVM_RBF, SVM_C_SVC
6
7 import cvutils, moving, ml
8
9
10 # todo update with argparse
11 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)
13 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)
15 parser.add_argument('-s', dest = 'rescaleSize', help = 'rescale size of image samples', default = 64, type = int)
16 parser.add_argument('-o', dest = 'nOrientations', help = 'number of orientations in HoG', default = 9, type = int)
17 parser.add_argument('-p', dest = 'nPixelsPerCell', help = 'number of pixels per cell', default = 8, type = int)
18 parser.add_argument('-c', dest = 'nCellsPerBlock', help = 'number of cells per block', default = 2, type = int)
19 args = parser.parse_args()
20
21 rescaleSize = (args.rescaleSize, args.rescaleSize)
22 nPixelsPerCell = (args.nPixelsPerCell, args.nPixelsPerCell)
23 nCellsPerBlock = (args.nCellsPerBlock, args.nCellsPerBlock)
24
25 imageDirectories = {'pedestrian': args.directoryName + "/Pedestrians/",
26 'bicycle': args.directoryName + "/Cyclists/",
27 'car': args.directoryName + "/Vehicles/"}
28
29 #directory_model = args.directoryName
30 trainingSamplesPBV = {}
31 trainingLabelsPBV = {}
32 trainingSamplesBV = {}
33 trainingLabelsBV = {}
34 trainingSamplesPB = {}
35 trainingLabelsPB = {}
36 trainingSamplesPV = {}
37 trainingLabelsPV = {}
38
39 for k, v in imageDirectories.iteritems():
40 print('Loading {} samples'.format(k))
41 trainingSamplesPBV[k], trainingLabelsPBV[k] = cvutils.createHOGTrainingSet(v, moving.userType2Num[k], rescaleSize, args.nOrientations, nPixelsPerCell, nCellsPerBlock)
42 if k != 'pedestrian':
43 trainingSamplesBV[k], trainingLabelsBV[k] = cvutils.createHOGTrainingSet(v, moving.userType2Num[k], rescaleSize, args.nOrientations, nPixelsPerCell, nCellsPerBlock)
44 if k != 'car':
45 trainingSamplesPB[k], trainingLabelsPB[k] = cvutils.createHOGTrainingSet(v, moving.userType2Num[k], rescaleSize, args.nOrientations, nPixelsPerCell, nCellsPerBlock)
46 if k != 'bicycle':
47 trainingSamplesPV[k], trainingLabelsPV[k] = cvutils.createHOGTrainingSet(v, moving.userType2Num[k], rescaleSize, args.nOrientations, nPixelsPerCell, nCellsPerBlock)
48
49 # Training the Support Vector Machine
50 print "Training Pedestrian-Cyclist-Vehicle Model"
51 model = ml.SVM(args.svmType, args.kernelType)
52 model.train(np.concatenate(trainingSamplesPBV.values()), np.concatenate(trainingLabelsPBV.values()))
53 model.save(args.directoryName + "/modelPBV.xml")
54
55 print "Training Cyclist-Vehicle Model"
56 model = ml.SVM(args.svmType, args.kernelType)
57 model.train(np.concatenate(trainingSamplesBV.values()), np.concatenate(trainingLabelsBV.values()))
58 model.save(args.directoryName + "/modelBV.xml")
59
60 print "Training Pedestrian-Cyclist Model"
61 model = ml.SVM(args.svmType, args.kernelType)
62 model.train(np.concatenate(trainingSamplesPB.values()), np.concatenate(trainingLabelsPB.values()))
63 model.save(args.directoryName + "/modelPB.xml")
64
65 print "Training Pedestrian-Vehicle Model"
66 model = ml.SVM(args.svmType, args.kernelType)
67 model.train(np.concatenate(trainingSamplesPV.values()), np.concatenate(trainingLabelsPV.values()))
68 model.save(args.directoryName + "/modelPV.xml")