input image 128x128
if not padding image size = 127x127 after ConvolutionForward
if padding 1,1 image size = 128x128

padding pod size kernel.
if kernel 1x1 then padding 0,0
if kernel 3x3 then padding 1,1
if kernel 4x4 then padding 2,2

ConvolutionForward always return gray image because all colors mixed.

cudnnSetConvolution2dDescriptor,[convolution_descriptor],1,1,\ ;this padding h & w
				1,1,\ ;uv stride. step kernel filter through the image
                                             1,1,\ ;upscaleXY error if 2,2
                                             CUDNN_CROSS_CORRELATION,CUDNN_DATA_FLOAT