According to
"Accelerating the Super-Resolution Convolutional Neural Network", ECCV2016
the learned deconvolution kernels are better than a single bicubic kernel. (Sec. 3.3)
//---------------------------------------------
// testing - set5 (scale x3) , training - rgb91
//----------------------------------------------
// bicubic - 32.83 dB | deconv - 32.95dB
//----------------------------------------------
According to "Accelerating the Super-Resolution Convolutional Neural Network", ECCV2016
the learned deconvolution kernels are better than a single bicubic kernel. (Sec. 3.3) //--------------------------------------------- // testing - set5 (scale x3) , training - rgb91 //---------------------------------------------- // bicubic - 32.83 dB | deconv - 32.95dB //----------------------------------------------