Keras pre trained models with theano backend

I would like transefer some pre-trained keras theano backend but i failed i have the pretrained models from this link

https://jayanthkoushik.github.io/neural_style.html

I have tried to transfer the back end to tensorflow :-

here is code

def get_transformer_net(X, weights=None):
    input_ = Input(tensor=X, shape=(3, 256, 256))
    y = conv_layer(input_, 32, 9)
    y = conv_layer(y, 64, 3, subsample=2)
    y = conv_layer(y, 128, 3, subsample=2)
    y = residual_block(y)
    y = residual_block(y)
    y = residual_block(y)
    y = residual_block(y)
    y = residual_block(y)
    y = conv_layer(y, 64, 3, upsample=2)
    y = conv_layer(y, 32, 3, upsample=2)
    y = conv_layer(y, 3, 9, only_conv=True)
    y = Activation("tanh")(y)
    y = Lambda(lambda x: x * 150, output_shape=(3, None, None))(y)

    net = Model(input=input_, output=y)
    if weights is not None:
        try:
            net.load_weights(weights)
        except OSError as e:
            print(e)
            sys.exit(1)
    return net

this is my conversation code :

# Build transformer model.

X = theano.shared(np.array([[[[]]]

is there any solution to transfer theano backend preconfigured models to coreml

any suggestion , and thanks in advance