model.py

from keras.layers import Convolution2D, Input
from keras.layers.core import Dense, Flatten, Lambda, Activation, Dropout
from keras.models import Model, Sequential
from keras.optimizers import SGD

import tensorflow as tf

from keras import backend as K
K.set_image_dim_ordering('tf')

def atan_layer(x):
    print(x, tf.mul(tf.atan(x), 2))
    return tf.mul(tf.atan(x), 2)

def atan_layer_shape(input_shape):
    return input_shape

def atan(x):
    return tf.atan(x)

def NVIDA():

    inputs = Input(shape=(66, 200, 3))
    conv_1 = Convolution2D(24, 5, 5, activation='relu', name='conv_1', subsample=(2, 2))(inputs)
    conv_2 = Convolution2D(36, 5, 5, activation='relu', name='conv_2', subsample=(2, 2))(conv_1)
    conv_3 = Convolution2D(48, 5, 5, activation='relu', name='conv_3', subsample=(2, 2))(conv_2)
    conv_3 = Dropout(.5)(conv_3)

    conv_4 = Convolution2D(64, 3, 3, activation='relu', name='conv_4', subsample=(1, 1))(conv_3)
    conv_5 = Convolution2D(64, 3, 3, activation='relu', name='conv_5', subsample=(1, 1))(conv_4)

    flat = Flatten()(conv_5)

    dense_1 = Dense(1164)(flat)
    dense_1 = Dropout(.5)(flat)
    dense_2 = Dense(100, activation='relu')(dense_1)
    dense_2 = Dropout(.5)(flat)
    dense_3 = Dense(50, activation='relu')(dense_2)
    dense_3 = Dropout(.5)(flat)
    dense_4 = Dense(10, activation='relu')(dense_3)
    dense_4 = Dropout(.5)(flat)

    final = Dense(1, activation=atan)(dense_4)
    #angle = Lambda(lambda x: tf.mul(tf.atan(x), 2))(final)


    model = Model(input=inputs, output=final)
    model.compile(
        optimizer=SGD(lr=.001, momentum=.9),
        loss='mse')

    return model