if K.image_dim_ordering() == 'th':
# Transpose image dimensions (Theano uses the channels as the 1st dimension)
im = im.transpose((2, 0, 1))
# Use pre-trained weights for Theano backend
weights_path = '../imagenet_models/densenet169_weights_th.h5'
else:
# Use pre-trained weights for Tensorflow backend
weights_path = '../imagenet_models/densenet169_weights_tf.h5'
# Insert a new dimension for the batch_size
im = np.expand_dims(im, axis=0)
# Test pretrained model
model = DenseNet(reduction=0.5, classes=1000, weights_path=weights_path)
sgd = SGD(lr=1e-2, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd,
loss='categorical_crossentropy',
metrics=['accuracy'])
out = model.predict(im)
# Load ImageNet classes file
classes = []
with open('resources/classes.txt', 'r') as list_:
for line in list_:
classes.append(line.rstrip('\n'))
print('Prediction: ' + str(classes[np.argmax(out)]))
del Y_train
del X_valid
del Y_valid
NumNonTrainable = [0,10,75,200,525]
X_testNP = np.zeros([len(X_test),224,224,3])
for id in range(0,len(X_testNP)):
im = cv2.resize(X_test[id], (224, 224))
X_testNP[id] = im
for ib in range(0,len(NumNonTrainable)):
# Test pretrained model
model = DenseNet(reduction=0.5, classes=classes, weights_path=weights_path, NumNonTrainable=NumNonTrainable[ib])
# Learning rate is changed to 1e-3
sgd = SGD(lr=1e-2, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd, loss='categorical_crossentropy', metrics=['accuracy'])
start = time.time()
model.fit_generator(generator=my_training_batch_generator,
epochs=5,
verbose=1,
shuffle=True,
validation_data=my_validation_batch_generator)
model.save_weights('/home/shayan/Codes/DenseNet-Keras-master/adversarial_weights_tf_' + str(ib) + '.h5')
im = cv2.imread(filename)
im = cv2.resize(im, (224, 224)).astype(np.float32)
# Subtract mean pixel and multiple by scaling constant
# Reference: https://github.com/shicai/DenseNet-Caffe
im[:, :, 0] = (im[:, :, 0] - 103.94) * 0.017
im[:, :, 1] = (im[:, :, 1] - 116.78) * 0.017
im[:, :, 2] = (im[:, :, 2] - 123.68) * 0.017
X_test.append(im)
Y_test.append([1])
X_test = np.array(X_test)
Y_test = np.array(Y_test)
Y_test = to_categorical(Y_test)
# Test pretrained model
model = DenseNet(reduction=0.5, classes=classes)
# Learning rate is changed to 1e-3
sgd = SGD(lr=1e-2, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd,
loss='categorical_crossentropy',
metrics=['accuracy'])
model.fit_generator(generator=my_training_batch_generator,
epochs=3,
verbose=1,
shuffle=True,
validation_data=my_validation_batch_generator)
f = open(
"/home/shayan/PycharmProjects/DenseNet-Keras-master/results/Stat_Results_NonTrainable_NonWeight"
for filename in glob.glob(
'/home/shayan/PycharmProjects/Dataset/Botnet/Test/' + str(i4) +
'/*.png'):
im = cv2.imread(filename)
im = cv2.resize(im, (224, 224)).astype(np.float32)
# Subtract mean pixel and multiple by scaling constant
# Reference: https://github.com/shicai/DenseNet-Caffe
im[:, :, 0] = (im[:, :, 0] - 103.94) * 0.017
im[:, :, 1] = (im[:, :, 1] - 116.78) * 0.017
im[:, :, 2] = (im[:, :, 2] - 123.68) * 0.017
X_test.append(im)
Y_test.append([1])
# Test pretrained model
model, logits = DenseNet(reduction=0.5,
classes=classes,
weights_path=weights_path)
# Learning rate is changed to 1e-3
sgd = SGD(lr=1e-2, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd,
loss='categorical_crossentropy',
metrics=['accuracy'])
class Dummy:
pass
env = Dummy()
if K.image_dim_ordering() == 'th':
# Transpose image dimensions (Theano uses the channels as the 1st dimension)
im = im.transpose((2,0,1))
# Use pre-trained weights for Theano backend
weights_path = 'imagenet_models/densenet169_weights_th.h5'
else:
# Use pre-trained weights for Tensorflow backend
weights_path = 'imagenet_models/densenet169_weights_tf.h5'
# Insert a new dimension for the batch_size
im = np.expand_dims(im, axis=0)
# Test pretrained model
model = DenseNet(reduction=0.5, classes=10, weights_path=weights_path)
# Learning rate is changed to 1e-3
sgd = SGD(lr=1e-2, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd, loss='categorical_crossentropy', metrics=['accuracy'])
out = model.predict(im)
# Load ImageNet classes file
classes = []
with open('resources/classes.txt', 'r') as list_:
for line in list_:
classes.append(line.rstrip('\n'))
print 'Prediction: '+str(classes[np.argmax(out)])
X_test.append(im)
Y_test.append([1])
X_test = np.array(X_test)
Y_test = np.array(Y_test)
Y_test = to_categorical(Y_test)
NumNonTrainable = [75]
print(X_test.shape)
print(Y_test.shape)
for ib in range(0,len(NumNonTrainable)):
# Test pretrained model
model = DenseNet(reduction=0.5, classes=classes, weights_path=weights_path, NumNonTrainable=NumNonTrainable[ib])
# Learning rate is changed to 1e-3
sgd = SGD(lr=1e-2, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd, loss='categorical_crossentropy', metrics=['accuracy'])
start = time.time()
model.fit_generator(generator=my_training_batch_generator,
epochs=1,
verbose=1,
shuffle=True,
validation_data=my_validation_batch_generator)
end = time.time()
train_time = end - start
if K.image_dim_ordering() == 'th':
# Transpose image dimensions (Theano uses the channels as the 1st dimension)
im = im.transpose((2, 0, 1))
# Use pre-trained weights for Theano backend
weights_path = 'imagenet_models/densenet169_weights_th.h5'
else:
# Use pre-trained weights for Tensorflow backend
weights_path = 'imagenet_models/densenet169_weights_tf.h5'
# Insert a new dimension for the batch_size
im = np.expand_dims(im, axis=0)
# Test pretrained model
model = DenseNet(reduction=0.5,
classes=classes,
weights_path=weights_path,
NumNonTrainable=10)
# Learning rate is changed to 1e-3
sgd = SGD(lr=1e-2, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd,
loss='categorical_crossentropy',
metrics=['accuracy'])
'''
out = model.predict(im)
# Load ImageNet classes file
classes = []
with open('resources/classes.txt', 'r') as list_:
for line in list_:
classes.append(line.rstrip('\n'))
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black")
plt.ylabel('True label')
plt.xlabel('Predicted label')
plt.tight_layout()
classes = 2
# Use pre-trained weights for Tensorflow backend
weights_path = 'imagenet_models/densenet169_weights_tf.h5'
# Test pretrained model
model = DenseNet(reduction=0.5,
classes=classes,
weights_path=weights_path,
NumNonTrainable=10)
# Learning rate is changed to 1e-3
sgd = SGD(lr=1e-2, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd,
loss='categorical_crossentropy',
metrics=['accuracy'])
print('Start Reading Data')
# "0" = Botnet Traffic Data and "1" = Normal Traffic Data
X_train = []
Y_train = []
for i1 in range(0, 12):
'cloudy': 6,
'conventional_mine': 7,
'cultivation': 8,
'habitation': 9,
'haze': 10,
'partly_cloudy': 11,
'primary': 12,
'road': 13,
'selective_logging': 14,
'slash_burn': 15,
'water': 16
}
inv_label_map = {i: l for l, i in label_map.items()}
base_model = DenseNet(
weights_path=
"/mnt/home/dunan/Learn/Kaggle/planet_amazon/pretrained_weights/densenet169_weights_tf.h5"
)
x = base_model.get_layer('relu').output
x_newfc = GlobalAveragePooling2D(name='final_pool')(x)
x_newfc = Dense(n_classes, name='fc6')(x_newfc)
x_newfc = Activation('softmax', name='prob')(x_newfc)
model = Model(inputs=base_model.input, outputs=x_newfc)
sgd = SGD(lr=1e-3, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd,
loss='categorical_crossentropy',
metrics=['accuracy'])
X_train = []
y_train = []