def get_model():
from donkeycar.parts.keras import Input, Convolution2D, Dropout, Dense, Flatten, Model
drop = 0.1
img_in = Input(shape=(256, 256, 3), name='img_in')
x = img_in
x = Convolution2D(24, (5, 5), strides=(2, 2), activation='relu', name="conv2d_1")(x)
x = Dropout(drop)(x)
x = Convolution2D(32, (5, 5), strides=(2, 2), activation='relu', name="conv2d_2")(x)
x = Dropout(drop)(x)
x = Convolution2D(64, (5, 5), strides=(2, 2), activation='relu', name="conv2d_3")(x)
x = Dropout(drop)(x)
x = Convolution2D(64, (3, 3), strides=(1, 1), activation='relu', name="conv2d_4")(x)
x = Dropout(drop)(x)
x = Convolution2D(64, (3, 3), strides=(1, 1), activation='relu', name="conv2d_5")(x)
x = Dropout(drop)(x)
x = Flatten(name='flattened')(x)
x = Dense(100, activation='relu')(x)
x = Dropout(drop)(x)
x = Dense(50, activation='relu')(x)
x = Dropout(drop)(x)
outputs = [
Dense(1, activation='linear', name='n_outputs0')(x)
]
# for i in range(2):
# outputs.append(Dense(1, activation='linear', name='n_outputs' + str(i))(x))
return Model(inputs=[img_in], outputs=outputs)
def get_small_epoch_model(input_shape=(66, 200, 3)):
if not input_shape:
input_shape = (66, 200, 3)
img_in = Input(shape=input_shape)
x = img_in
# Normalization
# x = Lambda(lambda x: x / 127.5 - 1.0)(x)
# x = Lambda(lambda x: x * 255 / 127.5 - 1.0)(x)
x = Convolution2D(16, (3, 3), activation='relu', padding='same')(x)
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = Dropout(0.25)(x)
x = Convolution2D(32, (3, 3), activation='relu', padding='same')(x)
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = Dropout(0.25)(x)
x = Convolution2D(64, (3, 3), activation='relu', padding='same')(x)
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = Dropout(0.5)(x)
x = Flatten()(x)
x = Dense(512, activation='relu')(x)
x = Dropout(.5)(x)
outputs = get_output_layers(x)
# for i in range(2):
# outputs.append(Dense(1, name='n_outputs' + str(i))(x))
return Model(inputs=[img_in], outputs=outputs)
def get_convolution_kernels(n, kernel_size):
return Convolution2D(n,
kernel_size,
kernel_initializer="he_normal",
bias_initializer="he_normal",
activation='relu',
padding='same')
def get_default_donkeycar_model(input_shape=(140, 320, 3)):
if not input_shape:
input_shape = (140, 320, 3)
drop = 0.1
img_in = Input(shape=input_shape, name='img_in')
x = img_in
x = Convolution2D(24, (5, 5),
strides=(2, 2),
activation='relu',
name="conv2d_1")(x)
x = Dropout(drop)(x)
x = Convolution2D(32, (5, 5),
strides=(2, 2),
activation='relu',
name="conv2d_2")(x)
x = Dropout(drop)(x)
x = Convolution2D(64, (5, 5),
strides=(2, 2),
activation='relu',
name="conv2d_3")(x)
x = Dropout(drop)(x)
x = Convolution2D(64, (3, 3),
strides=(1, 1),
activation='relu',
name="conv2d_4")(x)
x = Dropout(drop)(x)
x = Convolution2D(64, (3, 3),
strides=(1, 1),
activation='relu',
name="conv2d_5")(x)
x = Dropout(drop)(x)
x = Flatten(name='flattened')(x)
x = Dense(100, activation='relu')(x)
x = Dropout(drop)(x)
x = Dense(50, activation='relu')(x)
x = Dropout(drop)(x)
outputs = get_output_layers(x)
# for i in range(2):
# outputs.append(Dense(1, activation='linear', name='n_outputs' + str(i))(x))
return Model(inputs=[img_in], outputs=outputs)
def get_dave2_model(input_shape=(66, 200, 3)):
if not input_shape:
input_shape = (66, 200, 3)
img_in = Input(shape=input_shape, name='img_in')
x = img_in
# Normalization is carried out in donkey car train data generator
# x = Lambda(lambda x: x / 127.5 - 1., input_shape=input_shape, name='lambda_norm')(x)
x = Lambda(lambda x: x * 255 / 127.5 - 1.,
input_shape=input_shape,
name='lambda_norm')(x)
(66, 220)
# 5x5 Convolutional layers with stride of 2x2
x = Convolution2D(24, (5, 5),
strides=(2, 2),
name='conv1',
activation='elu')(x)
x = Convolution2D(36, (5, 5),
strides=(2, 2),
name='conv2',
activation='elu')(x)
x = Convolution2D(48, (5, 5),
strides=(2, 2),
name='conv3',
activation='elu')(x)
# 3x3 Convolutional layers with stride of 1x1
x = Convolution2D(64, (3, 3), name='conv4', activation='elu')(x)
x = Convolution2D(64, (3, 3), name='conv5', activation='elu')(x)
# Flatten before passing to Fully Connected layers
x = Flatten()(x)
# Three fully connected layers
x = Dense(100, name='fc1', activation='elu')(x)
x = Dense(50, name='fc2', activation='elu')(x)
x = Dense(10, name='fc3', activation='elu')(x)
# Output layer
outputs = get_output_layers(x)
return Model(inputs=[img_in], outputs=outputs)
def get_chaffeur_model(input_shape=(120, 320, 3)):
from donkeycar.parts.keras import SpatialDropout2D
if not input_shape:
input_shape = (120, 320, 3)
def get_convolution_kernels(n, kernel_size):
return Convolution2D(n,
kernel_size,
kernel_initializer="he_normal",
bias_initializer="he_normal",
activation='relu',
padding='same')
img_in = Input(shape=input_shape, name='img_in')
x = img_in
# (Convolution -> Spatial Dropout -> Max Pooling) x5
x = Convolution2D(16, (5, 5),
input_shape=input_shape,
kernel_initializer="he_normal",
bias_initializer="he_normal",
activation='relu',
padding='same')(x)
x = SpatialDropout2D(0.1)(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = get_convolution_kernels(20, (5, 5))(x)
x = SpatialDropout2D(0.1)(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = get_convolution_kernels(40, (3, 3))(x)
x = SpatialDropout2D(0.1)(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = get_convolution_kernels(60, (3, 3))(x)
x = SpatialDropout2D(0.1)(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = get_convolution_kernels(80, (2, 2))(x)
x = SpatialDropout2D(0.1)(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
# Flattening and dropping-out
x = Flatten()(x)
x = Dropout(0.5)(x)
# Classification
outputs = get_output_layers(x)
# for i in range(2):
# outputs.append(Dense(1, name='n_outputs' + str(i), kernel_initializer='he_normal')(x))
return Model(inputs=[img_in], outputs=outputs)