【PyTorch】教程：torch.nn.ConvTranspose2d

torch.nn.ConvTranspose2d

CLASS torch.nn.ConvTranspose2d(in_channels, out_channels, kernel_size, stride=1, padding=0, output_padding=0, groups=1, bias=True, dilation=1, padding_mode='zeros', device=None, dtype=None)

2D 反卷积；

该模块可以被视为 Conv2d 相对于其输入的梯度，也被称为分数跨步卷积或反卷积（尽管不是真正意义的反卷积，因为不计算卷积的真逆）。更多见 here 论文Deconvolutional Networks

参数

• in_channels ([int]) – 输入通道数
• out_channels ([int]) – 输出通道数
• kernel_size ([int] or [tuple]) – 卷积核大小
• stride ([int] or [tuple], optional ) – 步长，默认为1
• padding ([int] or [tuple], optional ) – zero-padding添加到输入的每个维度上，默认为0
• output_padding ([int] or [tuple], optional ) – 输出扩边，默认为0
• groups ([int], optional ) – 组别数，默认为1
• bias ([bool], optional ) – 如果为 True, 添加 bias 学习, 默认为 True
• dilation ([int], [tuple], optional ) – kernel 元素之间的空格数，默认为1

shape

• input: $(N,C_{in},H_{in},W_{in})$ 或 $(C_{in},H_{in},W_{in})$
• output: $(N,C_{out},H_{out},W_{out})$ 或 $(C_{out},H_{out},W_{out})$

其中

$$H_{out}=(H_{in}-1)\ast stride[0]-2\ast padding[0]+dilation[0]\ast (kernelsize[0]-1)+outputpadding[0]+1$$

$$W_{out}=(W_{in}-1)\ast stride[1]-2\ast padding[1]+dilation[1]\ast (kernelsize[1]-1)+outputpadding[1]+1$$

变量

• weight ([Tensor]) – 模块的可学习的权重参数，($\text{in\_channels},\frac{\text{out\_channels}}{\text{groups}},\text{kernel\_size[0]},\text{kernel\_size[1]})$. 这些权重值采样来自于 $\mathcal{U}(-\sqrt{k},\sqrt{k})$ 其中，$k=\frac{groups}{C_{\text{out}}\ast {\prod }_{i=0}^1\text{kernel\_size}[i]}$
• bias ([Tensor] ) – 模块的可学习的权重参数， $\mathcal{U}(-\sqrt{k},\sqrt{k})$, 其中 $k=\frac{groups}{C_{\text{out}}\ast {\prod }_{i=0}^1\text{kernel\_size}[i]}$

# torch.nn.ConvTranspose2d
import torch
import torch.nn as nn

# With square kernels and equal stride
m = nn.ConvTranspose2d(16, 33, 3, stride=2)
# non-square kernels and unequal stride and with padding
m = nn.ConvTranspose2d(16, 33, (3, 5), stride=(2, 1), padding=(4, 2))
input = torch.randn(20, 16, 50, 100)
output = m(input)
print("output.size: ", output.size()) # output.size:  torch.Size([20, 33, 93, 100])

# exact output size can be also specified as an argument
input = torch.randn(1, 16, 12, 12)
downsample = nn.Conv2d(16, 16, 3, stride=2, padding=1)
upsample = nn.ConvTranspose2d(16, 16, 3, stride=2, padding=1)
h = downsample(input)
print("h.size(): ", h.size()) # h.size():  torch.Size([1, 16, 6, 6])

output = upsample(h, output_size=input.size())
print("output.size(): ", output.size()) # output.size():  torch.Size([1, 16, 12, 12]) 

【参考】

ConvTranspose2d — PyTorch 1.13 documentation