Convolutional/Max pooling layer의 output size 계산식

Seohyun·2023년 8월 20일
딥러닝
0

정리

목록 보기
6/9
post-custom-banner

1D

1D conv

Output size=Input size−Kernel size+2∗PaddingStride+1Output\ size = \frac{Input \ size - Kernel \ size + 2 * Padding}{Stride} + 1

1D deconv

Output size=(Input size−1)∗Stride+Kernel size−2∗PaddingOutput \ size = (Input \ size - 1) * Stride + Kernel \ size - 2 * Padding

2D

2D conv

Output height=Input height−Kernel height+2∗Padding heightStride height+1Output \ height = \frac{Input \ height - Kernel \ height + 2 * Padding \ height}{Stride \ height} + 1
Outputwidth=Input width−Kernel width+2∗Padding widthStride width+1Output width = \frac{Input \ width - Kernel \ width + 2 * Padding \ width}{Stride \ width} + 1

2D deconv

Output height=(Input height−1)∗Stride height+Kernel height−2∗Padding heightOutput \ height = (Input \ height - 1) * Stride \ height + Kernel \ height - 2 * Padding \ height
Output width=(Input width−1)∗Stride width+Kernel width−2∗Padding widthOutput \ width = (Input \ width - 1) * Stride \ width + Kernel \ width - 2 * Padding \ width

2D Maxpooling

Output size=Input size−Pool sizeStride+1Output \ size = \frac{Input \ size - Pool \ size}{Stride} + 1

3D

3D conv

Output height=Input height−Kernel height+2∗Padding heightStride height+1Output \ height = \frac{Input \ height - Kernel \ height + 2 * Padding \ height}{Stride \ height} + 1
Output width=Input width−Kernel width+2∗Padding widthStride width+1Output \ width = \frac{Input \ width - Kernel \ width + 2 * Padding \ width}{Stride \ width} + 1
Output depth=Input depth−Kernel depth+2∗Padding depthStride depth+1Output \ depth = \frac{Input \ depth - Kernel \ depth + 2 * Padding \ depth}{Stride \ depth} + 1

3D deconv

Output depth=(Input depth−1)∗Stride depth+Kernel depth−2∗Padding depthOutput \ depth = (Input \ depth - 1) * Stride \ depth + Kernel \ depth - 2 * Padding \ depth
Output height=(Input height−1)∗Stride height+Kernel height−2∗Padding heightOutput \ height = (Input \ height - 1) * Stride \ height + Kernel \ height - 2 * Padding \ height
Output width=(Input width−1)∗Stride width+Kernel width−2∗Padding widthOutput \ width = (Input \ width - 1) * Stride \ width + Kernel \ width - 2 * Padding \ width
profile
Seohyun
Hail hamster
이전 포스트

Gaussian distribution (Normal distribution, 정규분포)

다음 포스트

Autoencoder

post-custom-banner

0개의 댓글