Conv2d#

classtorch.nn.modules.conv.Conv2d(in_channels,out_channels,kernel_size,stride=1,padding=0,dilation=1,groups=1,bias=True,padding_mode='zeros',device=None,dtype=None)[source]#

Applies a 2D convolution over an input signal composed of several inputplanes.

In the simplest case, the output value of the layer with input size$(N,C_{\text{in}},H,W)$(N,Cin​,H,W) and output$(N,C_{\text{out}},H_{\text{out}},W_{\text{out}})$(N,Cout​,Hout​,Wout​)can be precisely described as:

$$\text{out}(N_i,C_{{\text{out}}_j})=\text{bias}(C_{{\text{out}}_j})+\sum _{k=0}^{C_{\text{in}}-1}\text{weight}(C_{{\text{out}}_j},k)\star \text{input}(N_i,k)$$out(Ni​,Coutj​​)=bias(Coutj​​)+k=0∑Cin​−1​weight(Coutj​​,k)⋆input(Ni​,k)

where$\star$⋆ is the valid 2Dcross-correlation operator,$N$N is a batch size,$C$C denotes a number of channels,$H$H is a height of input planes in pixels, and$W$W iswidth in pixels.

This module supportsTensorFloat32.

On certain ROCm devices, when using float16 inputs this module will usedifferent precision for backward.

• stride controls the stride for the cross-correlation, a singlenumber or a tuple.

• padding controls the amount of padding applied to the input. Itcan be either a string {‘valid’, ‘same’} or an int / a tuple of ints giving theamount of implicit padding applied on both sides.

• dilation controls the spacing between the kernel points; alsoknown as the à trous algorithm. It is harder to describe, but thislinkhas a nice visualization of whatdilation does.

• groups controls the connections between inputs and outputs.in_channels andout_channels must both be divisible bygroups. For example,

  • At groups=1, all inputs are convolved to all outputs.

  • At groups=2, the operation becomes equivalent to having two convlayers side by side, each seeing half the input channelsand producing half the output channels, and both subsequentlyconcatenated.

  • At groups=in_channels, each input channel is convolved withits own set of filters (of size$\frac{\text{out\_channels}}{\text{in\_channels}}$in_channelsout_channels​).

The parameterskernel_size,stride,padding,dilation can either be:

• a singleint – in which case the same value is used for the height and width dimension

• atuple of two ints – in which case, the firstint is used for the height dimension,and the secondint for the width dimension

Note

Whengroups == in_channels andout_channels == K * in_channels,whereK is a positive integer, this operation is also known as a “depthwise convolution”.

In other words, for an input of size$(N,C_{in},L_{in})$(N,Cin​,Lin​),a depthwise convolution with a depthwise multiplierK can be performed with the arguments$(C_{\text{in}}=C_{\text{in}},C_{\text{out}}=C_{\text{in}}\times \text{K},\mathrm{.}\mathrm{.}\mathrm{.},\text{groups}=C_{\text{in}})$(Cin​=Cin​,Cout​=Cin​×K,...,groups=Cin​).

Note

In some circumstances when given tensors on a CUDA device and using CuDNN, this operator may select a nondeterministic algorithm to increase performance. If this is undesirable, you can try to make the operation deterministic (potentially at a performance cost) by settingtorch.backends.cudnn.deterministic=True. SeeReproducibility for more information.

Note

padding='valid' is the same as no padding.padding='same' padsthe input so the output has the shape as the input. However, this modedoesn’t support any stride values other than 1.

Note

This module supports complex data types i.e.complex32,complex64,complex128.

Parameters

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

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

• kernel_size (int ortuple) – Size of the convolving kernel

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

• padding (int,tuple orstr,optional) – Padding added to all four sides ofthe input. Default: 0

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

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

• bias (bool,optional) – IfTrue, adds a learnable bias to theoutput. Default:True

• padding_mode (str,optional) –'zeros','reflect','replicate' or'circular'. Default:'zeros'

Shape:

• Input:$(N,C_{in},H_{in},W_{in})$(N,Cin​,Hin​,Win​) or$(C_{in},H_{in},W_{in})$(Cin​,Hin​,Win​)

• Output:$(N,C_{out},H_{out},W_{out})$(N,Cout​,Hout​,Wout​) or$(C_{out},H_{out},W_{out})$(Cout​,Hout​,Wout​), where

  $$H_{out}=\lfloor \frac{H_{in}+2\times \text{padding}[0]-\text{dilation}[0]\times (\text{kernel\_size}[0]-1)-1}{\text{stride}[0]}+1\rfloor$$Hout​=⌊stride[0]Hin​+2×padding[0]−dilation[0]×(kernel_size[0]−1)−1​+1⌋
  $$W_{out}=\lfloor \frac{W_{in}+2\times \text{padding}[1]-\text{dilation}[1]\times (\text{kernel\_size}[1]-1)-1}{\text{stride}[1]}+1\rfloor$$Wout​=⌊stride[1]Win​+2×padding[1]−dilation[1]×(kernel_size[1]−1)−1​+1⌋

Variables

• weight (Tensor) – the learnable weights of the module of shape$(\text{out\_channels},\frac{\text{in\_channels}}{\text{groups}},$(out_channels,groupsin_channels​,$\text{kernel\_size[0]},\text{kernel\_size[1]})$kernel_size[0],kernel_size[1]).The values of these weights are sampled from$\mathcal{U}(-\sqrt{k},\sqrt{k})$U(−k​,k​) where$k=\frac{groups}{C_{\text{in}}\ast {\prod }_{i=0}^1\text{kernel\_size}[i]}$k=Cin​∗∏i=01​kernel_size[i]groups​

• bias (Tensor) – the learnable bias of the module of shape(out_channels). Ifbias isTrue,then the values of these weights aresampled from$\mathcal{U}(-\sqrt{k},\sqrt{k})$U(−k​,k​) where$k=\frac{groups}{C_{\text{in}}\ast {\prod }_{i=0}^1\text{kernel\_size}[i]}$k=Cin​∗∏i=01​kernel_size[i]groups​

Examples

>>># With square kernels and equal stride>>>m=nn.Conv2d(16,33,3,stride=2)>>># non-square kernels and unequal stride and with padding>>>m=nn.Conv2d(16,33,(3,5),stride=(2,1),padding=(4,2))>>># non-square kernels and unequal stride and with padding and dilation>>>m=nn.Conv2d(16,33,(3,5),stride=(2,1),padding=(4,2),dilation=(3,1))>>>input=torch.randn(20,16,50,100)>>>output=m(input)