Force usage of Theano's image ordering.

[ArthurEmidio]

• The method get_model() in train_steering_model.py provides the input image to the Lambda layer as a shape (ch, row, col). This is in the Theano image dimension format, instead of Tensorflow's, which is (row, col, ch).

• An alternative to this is to modify the keras.json file in the Keras installation path (usually located in ~/.keras/keras.json), in the image_dim_ordering row.