PyTorch网络搭建中*list的用法解析

Posted 2022-12-08 算法与编程之美

问题

stage1 = nn.Sequential(
    nn.Sequential(
        nn.Conv2d(16, 32, 3, 1, 1),
        nn.ReLU(),
        nn.Conv2d(32, 64, 3, 1, 1),
        nn.ReLU(),
    )

)
stage3 = nn.Sequential(
    nn.Conv2d(3, 16, 3, 1, 1),
    list(stage1.children())[0], #! 但是可以把list中的元素加进来
)

stage4 = nn.Sequential(
    nn.Conv2d(3, 16, 3, 1, 1),
    *list(stage1.children())[0], #! 解包序列后再将每个层加入进来
)

stage3和stage4都可以添加到nn.Sequential()中，二者的区别是什么？

方法

import torch
from torch import nn

stage1 = nn.Sequential(
    nn.Sequential(
        nn.Conv2d(16, 32, 3, 1, 1),
        nn.ReLU(),
        nn.Conv2d(32, 64, 3, 1, 1),
        nn.ReLU(),
    )

)

# stage2 = nn.Sequential(
#     nn.Conv2d(3, 16, 3, 1, 1),
#     list(stage1.children()), #! 不能把一个list加进来，因为list is not a Module subclass
# )
# print(stage2)

stage3 = nn.Sequential(
    nn.Conv2d(3, 16, 3, 1, 1),
    list(stage1.children())[0], #! 但是可以把list中的元素加进来
)
print(stage3)
'''stage3输出:
Sequential(
  (0): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (1): Sequential(
    (0): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU()
    (2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU()
  )
)
'''

'''stage3与stage4的主要区别是:
- stage3将整个Sequential加入进来,
- 而stage4首先将Sequential解包而后再加入进来;
'''
stage4 = nn.Sequential(
    nn.Conv2d(3, 16, 3, 1, 1),
    *list(stage1.children())[0], #! 解包序列后再将每个层加入进来
)
print(stage4)

''' stage4输出:
Sequential(
  (0): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (1): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (2): ReLU()
  (3): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (4): ReLU()
)
'''

结语