nn.Sequential()

最近在学习搭建VGG16模块时,发现使用nn.Sequential()一个不明白的点.代码如下

def vgg_block(conv_nums, in_channels, out_channels):
    block = []
    for i in range(conv_nums):
        if i == 0:
            block.append(nn.Conv2d(in_channels, out_channels, 3, 1, 1))
        else:
            block.append(nn.Conv2d(out_channels, out_channels, 3, 1, 1))
        block.append(nn.ReLU())
    # 每一个block块结束之后加上最大池化层
    block.append(nn.MaxPool2d(2, 2))
    return nn.Sequential(*block)

上面没有任何问题,然而当我改写成

def vgg_block(conv_nums, in_channels, out_channels):
    block = []
    net = nn.Sequential()
    for i in range(conv_nums):
        if i == 0:
            block.append(nn.Conv2d(in_channels, out_channels, 3, 1, 1))
        else:
            block.append(nn.Conv2d(out_channels, out_channels, 3, 1, 1))
        block.append(nn.ReLU())
    # 每一个block块结束之后加上最大池化层
    block.append(nn.MaxPool2d(2, 2))
    return net(*block)

到后面就会报错, python语法还是不牢固,这里net实例化之后,最后的net(*block)实际上已经开始调用forward()函数开始前向传播了,我却理解成了继续调用构造方法了.