function convnet2autoencoder(inNet)

    outNet = inNet:clone()

    for i=outNet:size(),1,-1 do

        if torch.typename(outNet:get(i)) == 'nn.View' then

            outNet:remove(i)

        elseif torch.typename(net:get(i)) == 'nn.Linear' then

            outNet:remove(i)

        elseif torch.typename(net:get(i)) == 'nn.LogSoftMax' then

            outNet:remove(i)

        end

    end

    for i=outNet:size(),1,-1 do

        if torch.typename(outNet:get(i)) == 'nn.SpatialMaxPooling' then

            local pool_layer = nn.SpatialMaxPooling(2,2,2,2)

            outNet:insert(pool_layer,i+1)

            outNet:remove(i)

            outNet:add(nn.SpatialMaxUnpooling(pool_layer))

        elseif torch.typename(outNet:get(i)) == 'nn.SpatialConvolution' then

            nInputPlane = outNet:get(i).nOutputPlane

            nOutputPlane = outNet:get(i).nInputPlane

            kW = outNet:get(i).kW

            kH = outNet:get(i).kH

            outNet:add(nn.SpatialFullConvolution(nInputPlane, nOutputPlane, kW, kH))

            outNet:add(nn.ReLU())

        end

    end

    return outNet

end

function autoencoder2convnet(net1, net2)

    -- get indices for convolution layers for net1

    convList1 = {}

    j = 1

    for i=1,net1:size() do

        if torch.typename(net1:get(i)) == 'nn.SpatialConvolution' then

            convList1[j] = i

            j = j + 1

        end

    end

    -- get indices for convolution layers for net2

    convList2 = {}

    j=1

    for i=1,net2:size() do

        if torch.typename(net2:get(i)) == 'nn.SpatialConvolution' then

            convList2[j] = i

            j = j + 1

        end

    end

    -- copy parameters from net1 to net2

    for i=1,#convList1 do

        net2:get(convList2[i]).weight = net1:get(convList1[i]).weight

        net2:get(convList2[i]).bias = net1:get(convList1[i]).bias

    end

    return net2

end