def x2Conv3x3BlockWithRes(layer,n,m):
    net = layer
    if (m != layer.output_shape[1]):
        layer = ConvLayer(layer, m, filter_size=1, pad=0, nonlinearity=None, b=None)
    net = ConvLayer(net, num_filters=n, filter_size=3, pad='same', nonlinearity=rectify, W=init.GlorotNormal())
    net = batch_norm(net)
    net = ConvLayer(net, num_filters=m, filter_size=3, pad='same', nonlinearity=None, W=init.GlorotNormal())
    net = ElemwiseSumLayer([net, layer])
    net = NonlinearityLayer(net, nonlinearity=rectify)
    net = PoolLayer(net, pool_size=(2,2), stride=(2,2), mode='max')
    return net

def build_model_5(X):
    net = InputLayer(shape=(None, 3, 32, 32), input_var=X)
    net = x2Conv3x3BlockWithRes(net, 64, 128)
    net = x2Conv3x3BlockWithRes(net, 96, 192)
    net = x2Conv3x3BlockWithRes(net, 128, 256)
    net = ConvLayer(net, num_filters=160, filter_size=3, pad='same', nonlinearity=rectify, W=init.GlorotNormal())
    net = batch_norm(net)
    net = ConvLayer(net, num_filters=320, filter_size=3, pad='same', W=init.GlorotNormal())
    net = batch_norm(net)
    net = PoolLayer(net, pool_size=(4,4),mode='max')
    net = DenseLayer(net, num_units=4096, nonlinearity=sigmoid,W=init.GlorotNormal())
    net = DenseLayer(net, num_units=10, nonlinearity=softmax,W=init.GlorotNormal())
    return net