def trainTopModelMobilenet():
print(
'\n*************************Train top models Mobilenet*****************************'
)
mobilenet = classes.ComplexModel(baseModels.Mobilenet(), (299, 299),
"mobilenet/")
data, labels, numClasses = mobilenet.loadAndPreprocessData(
destPathSelected, log=True)
numberSplits = classes.obtainNumSplitKFold(len(labels), 20)
folds = list(
StratifiedKFold(n_splits=numberSplits,
shuffle=True).split(data, labels))
for fold, (trainIdx, validIdx) in enumerate(folds):
print('\n[INFO] Executing fold ', fold)
trainData = data[trainIdx]
trainLabels = labels[trainIdx]
validData = data[validIdx]
validLabels = labels[validIdx]
mobilenet.buildTopModel(classes.modelType.typeA)
trainingTime, historyGenerated = mobilenet.trainTopModel(
trainData, trainLabels, validData, validLabels, fold)
mobilenet.testTopModel(validData, validLabels, trainingTime,
historyGenerated, fold)
print('\n************END ' + str(fold) + '***************')
def fineTuneMobilenet():
mobilenet = classes.ComplexModel(baseModels.Mobilenet(), (299, 299),
"mobilenet/")
data, labels, numClasses = mobilenet.loadAndPreprocessData(
destPathSelected, log=True)
numberSplits = classes.obtainNumSplitKFold(len(labels), 20)
folds = list(
StratifiedKFold(n_splits=numberSplits,
shuffle=True).split(data, labels))
informationBestModel = classes.obtainBestModel("mobilenet/topModel/")
print(
'\n*************************Fine Tune Mobilenet*****************************'
)
for fold, (trainIdx, validIdx) in enumerate(folds):
print('\n[INFO] Executing fold ', fold)
trainData = data[trainIdx]
trainLabels = labels[trainIdx]
validData = data[validIdx]
validLabels = labels[validIdx]
trainingTime, historyGenerated = mobilenet.fineTune(
trainData, trainLabels, validData, validLabels, 76,
informationBestModel[1], informationBestModel[0], fold)
mobilenet.testFineTuneModel(validData, validLabels, trainingTime,
historyGenerated, fold, 15)
print('\n************END ' + str(fold) + '***************')
def trainTopModelsVgg16():
# Necessary for split data
print(
'\n*************************Train top models VGG16*****************************'
)
vgg16Model = classes.Model(baseModels.VGG16(), (224, 224), "vgg16/")
data, labels, numClasses = vgg16Model.loadAndPreprocessData(
destPathSelected, log=True)
numberSplits = classes.obtainNumSplitKFold(len(labels), 20)
folds = list(
StratifiedKFold(n_splits=numberSplits,
shuffle=True).split(data, labels))
for fold, (trainIdx, validIdx) in enumerate(folds):
print('\n[INFO] Executing fold ', fold)
trainData = data[trainIdx]
trainLabels = labels[trainIdx]
validData = data[validIdx]
validLabels = labels[validIdx]
vgg16Model = classes.Model(baseModels.VGG16(), (224, 224), "vgg16/")
vgg16Model.numClasses = numClasses
vgg16Model.generateBottlenecks(trainData, trainLabels, validData,
validLabels, fold)
vgg16Model.buildTopModel(classes.modelType.typeA)
trainingTime, historyGenerated = vgg16Model.trainTopModel(
trainLabels, validLabels, fold)
vgg16Model.testTopModel(validData, validLabels, trainingTime,
historyGenerated, fold)
vgg16Model.buildTopModel(classes.modelType.typeB)
trainingTime, historyGenerated = vgg16Model.trainTopModel(
trainLabels, validLabels, fold)
vgg16Model.testTopModel(validData, validLabels, trainingTime,
historyGenerated, fold)
vgg16Model.buildTopModel(classes.modelType.typeC)
trainingTime, historyGenerated = vgg16Model.trainTopModel(
trainLabels, validLabels, fold)
vgg16Model.testTopModel(validData, validLabels, trainingTime,
historyGenerated, fold)
# Free memory
del vgg16Model
gc.collect()
K.clear_session()
print('\n************END ' + str(fold) + '***************')
def fineTuneVgg16():
print(
'\n*************************Fine Tune VGG16*****************************'
)
vgg16Model = classes.Model(baseModels.VGG16(), (224, 224), "vgg16/")
data, labels, numClasses = vgg16Model.loadAndPreprocessData(
destPathSelected, log=True)
numberSplits = classes.obtainNumSplitKFold(len(labels), 20)
folds = list(
StratifiedKFold(n_splits=numberSplits,
shuffle=True).split(data, labels))
informationBestModel = classes.obtainBestModel("vgg16/topModel/")
for fold, (trainIdx, validIdx) in enumerate(folds):
print('\n[INFO] Executing fold ', fold)
trainData = data[trainIdx]
trainLabels = labels[trainIdx]
validData = data[validIdx]
validLabels = labels[validIdx]
vgg16Model = classes.Model(baseModels.VGG16(), (224, 224), "vgg16/")
vgg16Model.numClasses = numClasses
vgg16Model.buildTopModel(informationBestModel[1])
trainingTime, historyGenerated = vgg16Model.fineTune(
trainData, trainLabels, validData, validLabels, 15,
informationBestModel[1], informationBestModel[0], fold)
vgg16Model.testFineTuneModel(validData, validLabels, trainingTime,
historyGenerated, fold, 15)
# Free memory
del vgg16Model
gc.collect()
K.clear_session()
#vgg16Model = classes.Model(baseModels.VGG16(), (224, 224), "vgg16/")
#vgg16Model.numClasses = numClasses
#vgg16Model.buildTopModel(informationBestModel[1])
#trainingTime, historyGenerated = vgg16Model.fineTune(trainData, trainLabels, validData, validLabels, 12, informationBestModel[1], informationBestModel[0], fold)
#vgg16Model.testFineTuneModel(validData, validLabels, trainingTime, historyGenerated, fold, 12)
print('\n************END Fold ' + str(fold) + '***************')