ConvTranspose2d

CLASStorch.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)

 

Parameters

• in_channels (int) – Number of channels in the input image

• out_channels (int) – Number of channels produced by the convolution

• kernel_size (int or tuple) – Size of the convolving kernel

• stride (int or tuple, optional) – Stride of the convolution. Default: 1

• padding (int or tuple, optional) – zero-padding will be added to both sides of each dimension in the input. Default: 0dilation * (kernel_size - 1) - padding

• output_padding (int or tuple, optional) – Additional size added to one side of each dimension in the output shape. Default: 0

• groups (int, optional) – Number of blocked connections from input channels to output channels. Default: 1

• bias (bool, optional) – If , adds a learnable bias to the output. Default: TrueTrue

• dilation (int or tuple, optional) – Spacing between kernel elements. Default: 1

shape:

input:(N,C_in,H_in,W_in) or (C_in,H_in,W_in) 

output:(N,C_out,H_out,W_out) or (C_out,H_out,W_out)

转置卷积运算步骤：

1、在输入特征图元素间填充s-1行、列0

2、在输入特征图四周填充k-p-1行、列0

3、将卷积核参数上下、左右翻转

4、做正常卷积运算（填充0，步距1）

H_out​ = (H_in​−1) × stride[0] − 2 × padding[0] + dilation[0] × (kernel_size[0] − 1) + output_padding[0] + 1

W_out​ = (W_in​−1) × stride[1] − 2 × padding[1] + dilation[1] × (kernel_size[1] − 1) + output_padding[1] + 1