What is your loss function? Also, is one deconvolution enough? Is there an advantage to using multiple deconvoluttion layers?