def createModel(self):
# Network structure must be directly changed here.
base_model = VGG16(backend=backend,
layers=layers,
models=models,
utils=utils,
weights='imagenet',
include_top=False,
input_shape=(self.img_rows, self.img_cols,
self.img_channels))
for layer in base_model.layers:
layer.trainable = False
x = base_model.output
# flat: passing it to a dense layer
x = Flatten()(x)
# 1st Dense Layer
x = Dense(4096, activation='relu', name='fc1')(x)
# 2st Dense Layer
x = Dense(4096, activation='relu', name='fc2')(x)
# Output Layer
x = Dense(network_outputs)(x)
model = Model(base_model.input, x)
#adam = Adam(lr=self.learningRate)
#model.compile(loss='mse',optimizer=adam)
model.compile(RMSprop(lr=self.learningRate), 'MSE')
model.summary()
return model
class CarClassifier:
pre_model = VGG16(weights=None,
include_top=False,
input_shape=(224, 224, 3),
backend=keras.backend,
layers=keras.layers,
models=keras.models,
utils=keras.utils)
pre_model.trainable = False
pre_model.summary()
# add output layer for VGG16 output (4096 -> 1000 => 4096 -> 1000 -> 100)
vgg_model = models.Sequential()
vgg_model.add(pre_model)
vgg_model.add(layers.Flatten())
vgg_model.add(layers.Dense(4096, activation='relu'))
vgg_model.add(layers.Dense(1024, activation='relu'))
vgg_model.add(layers.Dropout(0.5))
vgg_model.add(layers.Dense(20, activation='softmax')) # practical
vgg_model.summary()
vgg_model.compile(loss='categorical_crossentropy',
optimizer=optimizers.RMSprop(lr=2e-5),
metrics=['acc'])
output_label = vgg_model.predict_classes()
def __init__(self, network_arch='vgg'):
self.network_arch = network_arch
if network_arch == 'vgg':
self.model = VGG16(weights='imagenet', include_top=False)
else:
raise ValueError('Not defined for other values. Use only vgg')
self.model.summary()
def trenuj(self, x_train, x_test, y_train, y_test, il_etykiet):
self.il_etykiet = il_etykiet
#pobieramy parametry obecnej sieci neuronowej
il_neuronow = self.wybrane_parametry['neurony']
sposob_aktywacji = self.wybrane_parametry['aktywacja']
il_warstw = self.wybrane_parametry['il_warstw']
wejscie = Input(shape=(224, 224, 3))
#tworzymy model bazujac na VGG16
model = VGG16(input_tensor=wejscie,
include_top=False,
weights='imagenet')
x = model.get_layer('block5_pool').output
x = Flatten(name='flatten')(x)
# dodajemy warstwy
for i in range(il_warstw):
x = Dense(il_neuronow, activation=sposob_aktywacji)(x)
# warstwa wyjsciowa
wyjscie = Dense(self.il_etykiet, activation='softmax')(x)
nowy_model_vgg = Model(wejscie, wyjscie)
# zamrazamy warstwy poza nasza nowa, zeby nie szkolic od nowa
# orginalnego modelu
for warstwa in nowy_model_vgg.layers[:-(il_warstw + 1)]:
warstwa.trainable = False
nowy_model_vgg.compile(loss='categorical_crossentropy',
optimizer='rmsprop',
metrics=['accuracy'])
histogram = nowy_model_vgg.fit(x_train,
y_train,
batch_size=self.batch_size,
epochs=self.epochs,
verbose=self.verbose,
validation_data=(x_test, y_test))
# ewaluacja
(loss, accuracy) = nowy_model_vgg.evaluate(x_test,
y_test,
batch_size=self.batch_size,
verbose=self.verbose)
y_pred = nowy_model_vgg.predict(x_test, batch_size=self.batch_size)
self.confusion_matrixx = confusion_matrix(y_test.argmax(axis=-1),
y_pred.argmax(axis=-1))
self.histogram = histogram
self.strata = loss
self.celnosc = accuracy
def main():
im_root = tb.read_config('config.json', 'UCMerced_LandUse_directory')
# Choosing the number of layer from VGG
if VGG_LAYER == 16:
extractor = VGG16(weights='imagenet', include_top=False)
elif VGG_LAYER == 19:
extractor = VGG19(weights='imagenet', include_top=False)
# Get ImageData of 1 subsample (1 image every class)
# 'IMG' is a folder containing image data
data1 = getImageData('IMG/', 0, 21)
X = extractor.predict(data1, verbose=0)
# iterate for all 2100 images (100 times)
for i in range(1, 100):
data2 = getImageData('IMG/', i * 21, i * 21 + 21)
X2 = extractor.predict(data2, verbose=0)
X = np.concatenate((X, X2), axis=0)
print(X.shape) # to clarify only (optional)
# save the array to external file that will be used for Neural Network
np.save('Input_NN_VGG' + str(VGG_LAYER) + '.npy', X)
def load_model():
global network_model
network_model = VGG16(weights="imagenet")
global graph
graph = tf.get_default_graph()
def build(self, use_cpu=False, print_summary=False):
vgg16 = VGG16(weights="imagenet",
include_top=False,
input_shape=(224, 224, 3))
inputs = Input(shape=(224, 224, 3))
conv_block_1 = self.buildConv2DBlock(inputs, 64, 1, 2)
pool1, pool1_argmax = MaxPoolingWithArgmax2D()(conv_block_1)
conv_block_2 = self.buildConv2DBlock(pool1, 128, 2, 2)
pool2, pool2_argmax = MaxPoolingWithArgmax2D()(conv_block_2)
conv_block_3 = self.buildConv2DBlock(pool2, 256, 3, 3)
pool3, pool3_argmax = MaxPoolingWithArgmax2D()(conv_block_3)
conv_block_4 = self.buildConv2DBlock(pool3, 512, 4, 3)
pool4, pool4_argmax = MaxPoolingWithArgmax2D()(conv_block_4)
conv_block_5 = self.buildConv2DBlock(pool4, 512, 5, 3)
pool5, pool5_argmax = MaxPoolingWithArgmax2D()(conv_block_5)
fc6 = Conv2D(512, 7, use_bias=False, padding='valid',
name='fc6')(pool5) #4096
fc6 = BatchNormalization(name='batchnorm_fc6')(fc6)
fc6 = Activation('relu', name='relu_fc6')(fc6)
fc7 = Conv2D(512, 1, use_bias=False, padding='valid',
name='fc7')(fc6) #4096
fc7 = BatchNormalization(name='batchnorm_fc7')(fc7)
fc7 = Activation('relu', name='relu_fc7')(fc7)
x = Conv2DTranspose(512,
7,
use_bias=False,
padding='valid',
name='deconv-fc6')(fc7)
x = BatchNormalization(name='batchnorm_deconv-fc6')(x)
x = Activation('relu', name='relu_deconv-fc6')(x)
x = MaxUnpooling2D(name='unpool5')([x, pool4_argmax])
x.set_shape(conv_block_5.get_shape())
x = Conv2DTranspose(512,
3,
use_bias=False,
padding='same',
name='deconv5-1')(x)
x = BatchNormalization(name='batchnorm_deconv5-1')(x)
x = Activation('relu', name='relu_deconv5-1')(x)
x = Conv2DTranspose(512,
3,
use_bias=False,
padding='same',
name='deconv5-2')(x)
x = BatchNormalization(name='batchnorm_deconv5-2')(x)
x = Activation('relu', name='relu_deconv5-2')(x)
x = Conv2DTranspose(512,
3,
use_bias=False,
padding='same',
name='deconv5-3')(x)
x = BatchNormalization(name='batchnorm_deconv5-3')(x)
x = Activation('relu', name='relu_deconv5-3')(x)
x = MaxUnpooling2D(name='unpool4')([x, pool4_argmax])
x.set_shape(conv_block_4.get_shape())
x = Conv2DTranspose(512,
3,
use_bias=False,
padding='same',
name='deconv4-1')(x)
x = BatchNormalization(name='batchnorm_deconv4-1')(x)
x = Activation('relu', name='relu_deconv4-1')(x)
x = Conv2DTranspose(512,
3,
use_bias=False,
padding='same',
name='deconv4-2')(x)
x = BatchNormalization(name='batchnorm_deconv4-2')(x)
x = Activation('relu', name='relu_deconv4-2')(x)
x = Conv2DTranspose(256,
3,
use_bias=False,
padding='same',
name='deconv4-3')(x)
x = BatchNormalization(name='batchnorm_deconv4-3')(x)
x = Activation('relu', name='relu_deconv4-3')(x)
x = MaxUnpooling2D(name='unpool3')([x, pool3_argmax])
x.set_shape(conv_block_3.get_shape())
x = Conv2DTranspose(256,
3,
use_bias=False,
padding='same',
name='deconv3-1')(x)
x = BatchNormalization(name='batchnorm_deconv3-1')(x)
x = Activation('relu', name='relu_deconv3-1')(x)
x = Conv2DTranspose(256,
3,
use_bias=False,
padding='same',
name='deconv3-2')(x)
x = BatchNormalization(name='batchnorm_deconv3-2')(x)
x = Activation('relu', name='relu_deconv3-2')(x)
x = Conv2DTranspose(128,
3,
use_bias=False,
padding='same',
name='deconv3-3')(x)
x = BatchNormalization(name='batchnorm_deconv3-3')(x)
x = Activation('relu', name='relu_deconv3-3')(x)
x = MaxUnpooling2D(name='unpool2')([x, pool2_argmax])
x.set_shape(conv_block_2.get_shape())
x = Conv2DTranspose(128,
3,
use_bias=False,
padding='same',
name='deconv2-1')(x)
x = BatchNormalization(name='batchnorm_deconv2-1')(x)
x = Activation('relu', name='relu_deconv2-1')(x)
x = Conv2DTranspose(64,
3,
use_bias=False,
padding='same',
name='deconv2-2')(x)
x = BatchNormalization(name='batchnorm_deconv2-2')(x)
x = Activation('relu', name='relu_deconv2-2')(x)
x = MaxUnpooling2D(name='unpool1')([x, pool1_argmax])
x.set_shape(conv_block_1.get_shape())
x = Conv2DTranspose(64,
3,
use_bias=False,
padding='same',
name='deconv1-1')(x)
x = BatchNormalization(name='batchnorm_deconv1-1')(x)
x = Activation('relu', name='relu_deconv1-1')(x)
x = Conv2DTranspose(64,
3,
use_bias=False,
padding='same',
name='deconv1-2')(x)
x = BatchNormalization(name='batchnorm_deconv1-2')(x)
x = Activation('relu', name='relu_deconv1-2')(x)
output = Conv2DTranspose(21,
1,
activation='softmax',
padding='same',
name='output')(x)
model = Model(inputs=inputs, outputs=output)
vgg16 = VGG16(weights="imagenet",
include_top=False,
input_shape=(224, 224, 3))
self.model = model
if print_summary:
print(self.model.summary())
for layer in self.model.layers:
if layer.name.startswith('conv'):
block = layer.name[4:].split('-')[0]
depth = layer.name[4:].split('-')[1]
# apply vgg16 weights without bias
layer.set_weights([
vgg16.get_layer('block{}_conv{}'.format(
block, depth)).get_weights()[0]
])
self.model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy', 'mse'])
return False
except Exception as e:
print(e)
return False
if __name__ =="__main__":
baseline_01 = 0.55
baseline_02 = 1.0
writer = None
frame_per_second = 30
maxLost = 5 # maximum number of object losts counted when the object is being tracked
car_label_strlist = np.loadtxt("car_label2.csv", delimiter=',', dtype='str')
#load vgg16
pre_model = VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3), backend=keras.backend,
layers=keras.layers, models=keras.models, utils=keras.utils)
pre_model.trainable = True
pre_model.summary()
# add output layer for VGG16 output (4096 -> 1000 => 4096 -> 1000 -> 100)
vgg_model = models.Sequential()
vgg_model.add(pre_model)
vgg_model.add(layers.Flatten())
vgg_model.add(layers.Dense(4096, activation='relu'))
vgg_model.add(layers.Dense(1024, activation='relu'))
vgg_model.add(layers.Dense(20, activation='softmax')) # practical
vgg_model.summary()
vgg_model.compile(loss='categorical_crossentropy', optimizer=optimizers.RMSprop(lr=2e-5), metrics=['acc'])
#vgg_model.load_weights("y_s_weights.h5") # 가중치 모델 1
#vgg_model.load_weights("y_up_weights.h5") # 가중치 모델 2
# -*- coding: utf-8 -*-
"""vgg16_keras.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1kLOtfIlx8gpJamwqZSXZiJ67Tz7f1Pah
"""
import tensorflow
from keras_preprocessing.image import load_img, img_to_array
from keras_applications.vgg16 import preprocess_input, decode_predictions, VGG16
model = VGG16()
images = load_img("speaker.jpg", target_size=(224, 224))
img_array = img_to_array(images)
img_array
img_array = img_array.reshape(
(1, img_array.shape[0], img_array.shape[1], img_array.shape[2]))
img_array.shape
image = preprocess_input(img_array)
image
def train(version, batch_size, epochs):
img_rows = 512
img_cols = 512
channel = 3
num_of_train_samples = 70
num_of_test_samples = 19
train_data_path = 'F:\\PARAM\\DL\\diaretdb1_v_1_1\\diaretdb1_v_1_1\\resources\\images\\train'
test_data_path = 'F:\\PARAM\\DL\\diaretdb1_v_1_1\\diaretdb1_v_1_1\\resources\\images\\test'
train_datagen = ImageDataGenerator(rescale=1. / 255)
test_datagen = ImageDataGenerator(rescale=1. / 255)
train_generator = train_datagen.flow_from_directory(
train_data_path,
target_size=(img_rows, img_cols),
batch_size=batch_size,
class_mode='categorical')
validation_generator = test_datagen.flow_from_directory(
test_data_path,
target_size=(img_rows, img_cols),
batch_size=batch_size,
class_mode='categorical')
if version == 0:
model = VGG16(include_top=False,
weights='imagenet',
input_shape=(img_rows, img_cols, channel),
pooling='avg')
else:
model = VGG19(include_top=False,
weights='imagenet',
input_shape=(img_rows, img_cols, channel),
pooling='avg')
for layer in model.layers:
layer.trainable = False
x = model.output
predictions = Dense(2, activation='softmax')(x)
model = Model(input=model.input, output=predictions)
sgd = SGD(lr=1e-3, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd,
loss='categorical_crossentropy',
metrics=['accuracy'])
model.fit_generator(train_generator,
steps_per_epoch=num_of_train_samples // batch_size,
epochs=epochs,
verbose=1)
model.save('VGG_' + str(version) + 'Exudate.h5')
y_prediction = model.predict_generator(
validation_generator, num_of_test_samples // batch_size + 1)
y_pred = np.argmax(y_prediction, axis=1)
print('Confusion Matrix')
print(confusion_matrix(validation_generator.classes, y_pred))
print('Classification Report')
target_names = ['exudates', 'non_exudates']
print(
classification_report(validation_generator.classes,
y_pred,
target_names=target_names))
cnf_matrix = confusion_matrix(validation_generator.classes, y_pred)
plt.figure()
plot_confusion_matrix(cnf_matrix, classes=target_names)
plt.show()
return model
def Wls(video, start_time, location, latitude, longtitude):
""""""
savePath = video
video = 'static/videos/'+video
uploaded_time = 0.0
baseline_01 = 0.55
baseline_02 = 1.0
writer = None
frame_per_second = 30
maxLost = 3 # maximum number of object losts counted when the object is being tracked
car_label_strlist = np.loadtxt("Module/car_label2.csv", delimiter=',', dtype='str')
#load vgg16
pre_model = VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3), backend=keras.backend,
layers=keras.layers, models=keras.models, utils=keras.utils)
pre_model.trainable = True
pre_model.summary()
# add output layer for VGG16 output (4096 -> 1000 => 4096 -> 1000 -> 100)
vgg_model = models.Sequential()
vgg_model.add(pre_model)
vgg_model.add(layers.Flatten())
vgg_model.add(layers.Dense(4096, activation='relu'))
vgg_model.add(layers.Dense(1024, activation='relu'))
vgg_model.add(layers.Dense(20, activation='softmax')) # practical
vgg_model.summary()
vgg_model.compile(loss='categorical_crossentropy', optimizer=optimizers.RMSprop(lr=2e-5), metrics=['acc'])
#vgg_model.load_weights("y_s_weights.h5") # 가중치 모델 1
#vgg_model.load_weights("y_up_weights.h5") # 가중치 모델 2
vgg_model.load_weights("Module/CarDatabaseShare/y_seventh_car_weight.hdf5") #가중치 모델 3
# Load Yolo
net = cv.dnn.readNet("Module/CarDatabaseShare/yolov3.weights", "Module/CarDatabaseShare/yolov3.cfg")
classes = []
with open("Module/CarDatabaseShare/coco.names", "r") as f:
classes = [line.strip() for line in f.readlines()]
layer_names = net.getLayerNames()
output_layers = [layer_names[i[0] - 1] for i in net.getUnconnectedOutLayers()]
#print(len(classes))
colors = np.random.uniform(0, 255, size=(len(classes), 3))
#print(colors)
bbox_colors = np.random.randint(0, 255, size=(len(classes), 3))
#print(bbox_colors)
tracker = Tracker(maxLost=maxLost)
cap = cv.VideoCapture(video)
# Check if camera opened successfully
if not cap.isOpened():
print("Error opening video stream or file")
sys.exit()
vwidth = cap.get(3) # float
vheight = cap.get(4) # float
frame_count = 0
while True:
_, frame = cap.read()
if _ is not True:
break
height, width, channels = frame.shape
# Detecting objects
blob = cv.dnn.blobFromImage(frame, 0.00392, (320, 320), (0, 0, 0), True, crop=False)
net.setInput(blob)
outs = net.forward(output_layers)
# Showing informations on the screen
class_ids = []
confidences = []
boxes = []
detected_boxes = []
detected_label = []
detected_times = []
detected_colors = []
for out in outs:
for detection in out:
scores = detection[5:]
class_id = np.argmax(scores)
confidence = scores[class_id]
if confidence > 0.5:
# Object detected
center_x = int(detection[0] * width)
center_y = int(detection[1] * height)
w = int(detection[2] * width)
h = int(detection[3] * height)
# Rectangle coordinates
x = int(center_x - w / 2)
y = int(center_y - h / 2)
boxes.append([x, y, w, h])
confidences.append(float(confidence))
class_ids.append(class_id)
indexes = cv.dnn.NMSBoxes(boxes, confidences, 0.5, 0.4)
# print(indexes)
font = cv.FONT_HERSHEY_PLAIN
for i in range(len(boxes)):
if i in indexes:
x, y, w, h = boxes[i]
if h > 40 and w > 50 and class_ids[i] == 2 and y > (vheight * baseline_01) and (y + h) < (vheight * baseline_02):
cropimg = frame.copy()
cropimg = frame[y:y+h, x:x+w]
try:
colorlabel = process_image(cropimg)
cropimg = cv.cvtColor(cropimg, cv.COLOR_BGR2GRAY)
cropimg = cv.resize(cropimg, dsize=(224, 224), interpolation=cv.INTER_AREA)
cropimg = cv.GaussianBlur(cropimg, (3, 3), 0)
cannimg = cv.Canny(cropimg, 25, 50)
npimg = np.asarray(cannimg)
npimg = np.stack((cannimg,)*3, axis=-1)
npimg = np.expand_dims(npimg, axis=0)
#print(npimg.shape)
car_recog = CarClassifier(npimg, vgg_model, car_label_strlist)
detected_boxes.append((x, y, x + w, y + h))
detected_label.append(car_recog)
detected_times.append(int(frame_count/frame_per_second) + 1)
detected_colors.append(colorlabel)
except:
continue
color = colors[i]
cv.rectangle(frame, (x, y), (x + w, y + h), color, 2)
#cv.putText(frame, car_recog, (x, y + 30), font, 2, color, 3)
#cv.putText(frame, label, (x, y + 20), font, 3, color, 3)
elif (class_ids[i] == 3 or class_ids[i] == 5 or class_ids[i] == 7) and y > (vheight * baseline_01) and (y + h) < (vheight * baseline_02):
cropimg = frame.copy()
cropimg = frame[y:y+h, x:x+w]
colorlabel = process_image(cropimg)
label = str(classes[class_ids[i]])
color = colors[i]
cv.rectangle(frame, (x, y), (x + w, y + h), color, 2)
#cv.putText(frame, label, (x, y + 20), font, 2, color, 3)
detected_boxes.append((x, y, x + w, y + h))
detected_label.append(label)
detected_times.append(int(frame_count/frame_per_second) + 1)
detected_colors.append(colorlabel)
cv.line(frame, (0,int(vheight*baseline_01)), (int(vwidth-1), int(vheight*baseline_01)), (0,0,0), 2)
cv.line(frame, (0, int(vheight * baseline_02)), (int(vwidth - 1), int(vheight * baseline_02)), (0, 0, 0), 2)
objects = tracker.update(detected_boxes, detected_label, detected_times, detected_colors) # update tracker based on the newly detected objects
for (objectID, centroid) in objects.items():
text = "ID {}".format(objectID)
if centroid[1] > (vheight * baseline_01) and centroid[1] < (vheight * baseline_02):
cv.putText(frame, text, (centroid[0] - 10, centroid[1] - 10), cv.FONT_HERSHEY_SIMPLEX,
0.5, (0, 255, 0), 2)
cv.circle(frame, (centroid[0], centroid[1]), 4, (0, 255, 0), -1)
# cv.imshow('', frame)
key = cv.waitKey(1)
if key == 27:
break
if writer is None:
startTime = start_time.replace(" ", "-")
startTime = startTime.replace(":", "-")
savepath=savePath.split(".")[0]
filePath= "static/output/" + savepath + "-" + startTime + ".mp4"
print(filePath)
fourcc = cv.VideoWriter_fourcc(*"h264")
# fourcc = cv.VideoWriter_fourcc(*"mp4v")
writer = cv.VideoWriter(filePath, fourcc, 30, (int(vwidth), int(vheight)), True)
writer.write(frame)
if frame_count % 50 == 0:
classified_info_list = tracker.getClassifiedInfo()
for obj in classified_info_list:
print(obj)
frame_count+=1
cctv = Cctv(video_link=filePath, location=location, start_time=start_time, latitude=latitude, longtitude=longtitude)
cctv.save()
classified_info_list = tracker.getClassifiedInfo()
for obj in classified_info_list:
print(obj)
#car model 이름 나오게하기, appearance_time 에다가 초 더하기, 브랜드 넣으면 좋음
#CCTV video_link 제대로 넣기 // 했고
for obj in classified_info_list:
try:
model_brand = obj[1][0][0].split("-")
model = model_brand[0]
brand = model_brand[1]
except:
model = obj[1][0][0]
brand = obj[1][0][0]
try:
car = Car.objects.get(model=model)
# car = Car.objects.get(model='a')
except:
car = Car(model=model, brand=brand)
# car = Car(model='a', brand="")
car.save()
start_time = datetime.datetime.strptime(str(start_time), '%Y-%m-%d %H:%M:%S')
appearance_time = start_time + timedelta(seconds=obj[2])
cctv_log = CctvLog(car_model=car, color=obj[3], appearance_time=appearance_time, cctv_id=cctv)
cctv_log.save()
writer.release()
cap.release()
os.remove(video)
cv.destroyAllWindows()
def set_net(self):
"""
build encoder-decoder network using the (pre-trained) VGG16 network.
:return: None
"""
# make input tensor based on numpy array
inp = Input(shape=(*self.img_size, 3), name='image')
# pre-trained encoder (vgg16)
encoder = VGG16(include_top=False, input_tensor=inp)
# first decoding block
decoder = UpSampling2D()(encoder.get_layer('block5_conv3').output)
if self.skip_connections:
decoder = concatenate(
[decoder, encoder.get_layer('block4_conv3').output], axis=-1)
decoder = Conv2D(256, (3, 3), activation='relu',
padding='same')(decoder)
decoder = Conv2D(256, (3, 3), activation='relu',
padding='same')(decoder)
# second decoding block
decoder = UpSampling2D()(decoder)
if self.skip_connections:
decoder = concatenate(
[decoder, encoder.get_layer('block3_conv3').output], axis=-1)
decoder = Conv2D(128, (3, 3), activation='relu',
padding='same')(decoder)
decoder = Conv2D(128, (3, 3), activation='relu',
padding='same')(decoder)
# third decoding block
decoder = UpSampling2D()(decoder)
if self.skip_connections:
decoder = concatenate(
[decoder, encoder.get_layer('block2_conv2').output], axis=-1)
decoder = Conv2D(64, (3, 3), activation='relu',
padding='same')(decoder)
decoder = Conv2D(64, (3, 3), activation='relu',
padding='same')(decoder)
# fourth decoding block
decoder = UpSampling2D()(decoder)
if self.skip_connections:
decoder = concatenate(
[decoder, encoder.get_layer('block1_conv2').output], axis=-1)
decoder = Conv2D(32, (3, 3), activation='relu',
padding='same')(decoder)
decoder = Conv2D(32, (3, 3), activation='relu',
padding='same')(decoder)
# tensor with the output mask
res_mask = Conv2D(self.mask_channels, (1, 1),
activation='softmax',
name='mask')(decoder)
if self.classification:
# build result tensor based on the output of the encoder
res_classification = self._add_classification_branch(encoder)
# model for multiple outputs (including the classification)
model = Model([inp], [res_mask, res_classification])
else:
# model for single output (only the mask)
model = Model([inp], [res_mask])
# store model and meta-information regarding the layer names, shape, and index of the encoder layers
self.neural_net = model
self.encoder_layers = [(layer.name, layer.output_shape, i)
for i, layer in enumerate(encoder.layers)]
Y = np_utils.to_categorical(labels, num_classes)
#Shuffle the dataset
x, y = shuffle(img_data, Y, random_state=2)
# Split the dataset
X_train, X_test, y_train, y_test = train_test_split(x,
y,
test_size=0.2,
random_state=2)
#########################################################################################
# Custom_vgg_model_1
#Training the classifier alone
image_input = Input(shape=(224, 224, 3))
model = VGG16(input_tensor=image_input, include_top=True, weights='imagenet')
model.summary()
last_layer = model.get_layer('fc2').output
#x= Flatten(name='flatten')(last_layer)
out = Dense(num_classes, activation='softmax', name='output')(last_layer)
custom_vgg_model = Model(image_input, out)
custom_vgg_model.summary()
for layer in custom_vgg_model.layers[:-1]:
layer.trainable = False
custom_vgg_model.layers[3].trainable
custom_vgg_model.compile(loss='categorical_crossentropy',
optimizer='rmsprop',
metrics=['accuracy'])
