torch.nn.Conv2d()使用

API

输入：[ batch_size, channels, height_1, width_1 ]

Conv2d输入参数：[ channels, output, height_2, width_2 ]

输出：[ batch_size,output, height_3, width_3 ]

实例：

def torch_practice():
    x = torch.randn(2,1,16,4)
    conv = torch.nn.Conv2d(1, 32, (2,2))
    res = conv(x)
    print(res.shape)

if __name__ == '__main__':
    torch_practice()

输出：torch.Size([2, 32, 15, 3])

batch大小不变：2

输出通道加厚：32。由卷积核的通道数决定

卷积结果:[15,3]。计算公式，n-m+1, 16-2+1=15