卷积尺寸计算公式（速查备用）

torch代码计算

def paras_cnn(k,s,p,i=64):
    x=torch.ones(1,1,i,i)
    conv = torch.nn.Conv2d(1, 1, kernel_size=k, stride=s, padding=p)
    convt= torch.nn.ConvTranspose2d(1, 1, kernel_size=k, stride=s, padding=p)
    h1=conv(x)
    h2=convt(x)
    y=convt(h1)
    print("conv(x):{} \t convT(x):{} \t convT(conv(x)):{}".format((h1.shape[2],h1.shape[3]),(h2.shape[2],h2.shape[3]),(y.shape[2],y.shape[3])))
    return h1.shape[2],h1.shape[3],h2.shape[2],h2.shape[3],y.shape[2],y.shape[3]