1.nn.Convv2d
from PIL import Image
from torchvision.transforms import ToTensor, ToPILImage
to_tensor = ToTensor() # img ->tensor
to_pil = ToPILImage() # tensor -> image
ll = Image.open('imgs/lena.png')
input = to_tensor(lena).unsqueeze(0) # 将batch size 设置为1
# conv = nn.Conv2d(in_channels, out_channels, \
kernel_size, stride=1,padding=0, dilation=1, groups=1, bias=True))
conv = nn.Conv2d(1,1,(3,3),1,bias=Flase)
2. AvgPool
pool = nn.AvgPool2d(2,2)
out = pool( V(input) )
3. Linear 全连接
input = V(t.randn(2,3))
linear = nn.Linear(3,4)
h = linear(input)
4. 激活函数
relu = nn.ReLU(inplace=True)
output = relu(input)
ReLU函数有个inplace参数，如果设为True,它会把输出直接覆盖到输入中，这样可以节省内反向传播的梯度。但是只有少数的autograd
