参考:https://pytorch.org/docs/stable/generated/torch.nn.PixelShuffle.html
pixel_shuffle = nn.PixelShuffle(3)
input = torch.randn(1, 9, 4, 4)
output = pixel_shuffle(input)
print(output.size()) # 1, 1,12,12
In [10]: input.sum()
Out[10]: tensor(-12.1099)
In [11]: output.sum()
Out[11]: tensor(-12.1099)
将C维度分配到HW上,进行resize操作。并不会产生新的元素(如果整除的话)