nn.Conv2d(int_c, out_c, kernel-size=(3,3), stride = (1,1), padding=(1,1))
nn.BatchNorm2d(num_feature) #特征图数目
F.avg_pool2d( input_data, kernel-size)