Pytorch-nn.ConvTransposed2d()
ConvTransposed2d()其实是Conv2d()的逆过程,其参数是一样的
Conv2d():
output = (input+2*Padding-kernelSize) / stride + 1(暂时不考虑outputPadding 注意:outputPadding只是在一边Padding)
=>input = (output-1) * stride - 2*Padding + kernelSize
例如输入图片尺寸为128,inputPadding为0,kernelSize为4,stride为2,outputPadding为1,那么输出图片尺寸为64
同理可得
ConvTransposed2d():
output = (input-1) * stride - 2*Padding + kernelSize