【PyTorch】教程：torch.nn.maxpool2d

torch.nn.MaxPool2d

CLASS torch.nn.MaxPool2d(kernel_size, stride=None, padding=0, dilation=1, return_indices=False, ceil_mode=False)

2D 最大池化；
输入 $(N,C,H,W)$, 输出 $(N,C,H_{out},W_{out})$ kernel_size $(kH,kW)$

$\begin{array}{rl}out(N_i,C_j,h,w)=& \underset{m=0,\dots ,kH-1}{\max }\underset{n=0,\dots ,kW-1}{\max }\\ & \text{input}(N_i,C_j,\text{stride[0]}\times h+m,\text{stride[1]}\times w+n)\end{array}$

参数

• kernel_size (Union [ int, Tuple [int, int]] ) – 池化窗口大小
• stride (Union [ int, Tuple [int, int]]) – 滑动步长，默认为 kernel_size
• padding (Union [ int, Tuple [int, int]]) – 扩边
• dilation (Union [ int, Tuple [int, int]]) – 控制窗口中元素的步长
• return_indices ([bool]) - 如果为 True, 返回最大值和索引；
• ceil_mode ([bool]) – 如果为 True, 使用 ceil 代替 floor 计算输出形状

形状

• Input: $(N,C,H_{in},W_{in})$ or $(C,H_{in},W_{in})$
• Output: $(N,C,H_{out},W_{out})$ or $(C,H_{out},W_{out})$ 其中
  $H_{out}=\lfloor \frac{H_{in}+2\ast \text{padding[0]}-\text{dilation[0]}\times (\text{kernel\_size[0]}-1)-1}{\text{stride[0]}}+1\rfloor$
  $W_{out}=\lfloor \frac{W_{in}+2\ast \text{padding[1]}-\text{dilation[1]}\times (\text{kernel\_size[1]}-1)-1}{\text{stride[1]}}+1\rfloor$

示例

# pool of square window of size=3, stride=2
m = nn.MaxPool2d(3, stride=2)
# pool of non-square window
m = nn.MaxPool2d((3, 2), stride=(2, 1))
input = torch.randn(20, 16, 50, 32)
output = m(input)
print(output.size()) # torch.Size([20, 16, 24, 31])

【参考】

MaxPool2d — PyTorch 1.13 documentation