딥러닝에서 사용되는 다양한 Convolution 기법들

[참고 사이트]

https://github.com/vdumoulin/conv_arithmetic/blob/master/README.md

GitHub - vdumoulin/conv_arithmetic: A technical report on convolution arithmetic in the context of deep learning

https://eehoeskrap.tistory.com/431

[Deep Learning] 딥러닝에서 사용되는 다양한 Convolution 기법들

https://hichoe95.tistory.com/48

Different types of Convolutions (Grouped convolution, depthwise convolution, pointwise convolution, depthwise separable convolut

https://coding-yoon.tistory.com/122

[딥러닝] Depth-wise Separable Convolution 원리(Pytorch 구현)

 

 

1. 기본 Convolution

<설명>

<example>

class torch.nn.Conv2d(in_channels

,

out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True, padding_mode='zeros', device=None, dtype=None)

>>> # 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)

https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html?highlight=dilated%20convolution 

Conv2d — PyTorch 1.9.0 documentation

2. Dilated Convolution

저번에 dilated UNet 만들때 끝단 stage 부분에서 사용된 convolution

3. Transposed Convolution

 

 

4. Separable Convolution

 

5. Depthwise Convolution

<example>

class depthwise_conv(nn.Module):
    def __init__(self, nin, kernels_per_layer):
        super(depthwise_separable_conv, self).__init__()
        self.depthwise = nn.Conv2d(nin, nin * kernels_per_layer, kernel_size=3, padding=1, groups=nin)


    def forward(self, x):
        out = self.depthwise(x)
        return out

첫 칸은 in_channels 이고, 두번째 칸은 out_channels

하나의 레이어에서 몇 개의 아웃풋을 만들 것이냐는 뜻

여기에서의 구현의 핵심은 groups = nin 부분인데, 입력 레이어를 nin개의 서로 다른 group으로 만들어서 해당 연산을 수행하겠다는 뜻이 된다. (반대로 생각해보면, group의 default 값이 1이라는 소리이다.)

 

6. Depthwise Separable Convolution (경량화 하는 방법)

 

Depthwise separable convolution = Depthwise conv + Pointwise conv

 

# of Parameters

 

단순히 Depthwise conv후에 Pointwise conv를 수행하는 방식이므로 두 방식의 param을 더해주기만 하면 된다.

따라서 총 param수는 K2C + CM = C(K2 + M)이 된다.

 

<example>

class depthwise_separable_conv(nn.Module):
    def __init__(self, nin, kernels_per_layer, nout):
        super(depthwise_separable_conv, self).__init__()
        self.depthwise = nn.Conv2d(nin, nin * kernels_per_layer, kernel_size=3, padding=1, groups=nin)
        self.pointwise = nn.Conv2d(nin * kernels_per_layer, nout, kernel_size=1)

    def forward(self, x):
        out = self.depthwise(x)
        out = self.pointwise(out)
        return out

<참고 사이트>

https://discuss.pytorch.org/t/how-to-modify-a-conv2d-to-depthwise-separable-convolution/15843/7

How to modify a Conv2d to Depthwise Separable Convolution?

 

7. Pointwise Convolution

<example>

class pointwise_conv(nn.Module):
    def __init__(self, nin, nout):
        super(depthwise_separable_conv, self).__init__()
        self.pointwise = nn.Conv2d(nin, nout, kernel_size=1)

    def forward(self, x):
        out = self.pointwise(x)
        return out

 

8. Grouped Convolution

 

9. Defornable Convolution