def mLeNetDefault_Features_Dense(input_shape,
num_classes,
steps_per_epoch,
epochs,
use_cache=False,
dataset='dataset1'):
model_name = 'LeNetDefault_Features_Dense'
helpers.createFoldersForModel(model_name, dataset)
print("===================== " + model_name +
" model ====================")
losses = {
'classifierOutput': classifierLossFunction,
'featuresOutput': featuresLossFunction
}
metrics = {
'classifierOutput': classifierAccuracy,
'featuresOutput': featuresAccuracy
}
weights = {'classifierOutput': 1.0, 'featuresOutput': 1.0}
if existsModelCache(model_name, dataset) and use_cache:
model = loadModel(model_name, dataset)
model.compile(loss=losses,
metrics=metrics,
optimizer='rmsprop',
run_eagerly=True,
loss_weights=weights)
evaluateModel(model, dataset, input_shape, steps_per_epoch, epochs,
True)
return model
model = getDefaultModel(input_shape)
dense1 = Dense(120)(model.layers[-1].output)
dense2 = Dense(84)(dense1)
dense3 = Dense(units=64, activation="relu")(dense2)
classifierOutput = Dense(num_classes,
activation='softmax',
name='classifierOutput')(dense3)
featuresOutput = Flatten(name='featuresOutput')(model.layers[-1].output)
# define new model
model = Model(inputs=model.inputs,
outputs=[classifierOutput, featuresOutput])
model.compile(loss=losses,
metrics=metrics,
optimizer='rmsprop',
run_eagerly=True,
loss_weights=weights)
model.run_eagerly = True
[history, time_taken] = fitModel(model, dataset, input_shape,
steps_per_epoch, epochs, True)
if use_cache:
saveModel(model, model_name, dataset)
createAndSaveCurvesFeatures(history, model_name, dataset)
return model
def mLeNetDefault(input_shape,
num_classes,
steps_per_epoch,
epochs,
use_cache=False,
dataset='dataset1'):
model_name = 'LeNetDefault'
helpers.createFoldersForModel(model_name, dataset)
print("===================== " + model_name +
" model ====================")
if existsModelCache(model_name, dataset) and use_cache:
model = loadModel(model_name, dataset)
model.compile(loss='categorical_crossentropy',
optimizer='rmsprop',
metrics=['accuracy'])
evaluateModel(model, dataset, input_shape, steps_per_epoch, epochs)
return model
model = getDefaultModel(input_shape)
model.add(Dense(120))
model.add(Dense(84))
model.add(Dense(num_classes, activation='softmax'))
# model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy',
optimizer='rmsprop',
metrics=['accuracy'])
[history, time_taken] = fitModel(model, dataset, input_shape,
steps_per_epoch, epochs, False)
if use_cache:
saveModel(model, model_name, dataset)
createAndSaveCurves(history, model_name, dataset)
return model