필터 지우기
필터 지우기

why resnet-18 has 71x1 layers?

조회 수: 11 (최근 30일)
TAN HOR YAN
TAN HOR YAN 2021년 11월 11일
답변: Srivardhan Gadila 2021년 11월 16일
Resnet-18 suppose has 18 later but why run in MATLAB it shows the DAGNetworks:
Layers: 71x1 layers
Connection: 78 x 2 tables
What does this properties means and why resnet 18 has 71 layers rather than 18 layers?

답변 (1개)

Srivardhan Gadila
Srivardhan Gadila 2021년 11월 16일
As per my knowledge, the number 18 refers to the core/main/important layers the architecture is based on i.e., the convolution and fullyconnected layers. In vgg16 it would be 16 and in vgg19 it would be 19 etc.
You can use analyzeNetwork function to analyze the network architecture. So if you check the count of only the convolution and fullyconnected layers in the resnet18 pretrained network, it will be 21, but in these the layers with names "res3a_branch1", "res4a_branch1" and "res5a_branch1" are 1x1 convolution layers which are used to control the depth so that the skip connections are possible accross the branches. Hence removing these 3 would leave us with 18. You can check the same with other variants of resnet as well (resnet50, resnet101 etc).
Fore more information you can refer to the References mentioned in the documentation page of resnet18. For list of deep learning layers in MATLAB refer to List of Deep Learning Layers.

카테고리

Help CenterFile Exchange에서 Image Data Workflows에 대해 자세히 알아보기

제품


릴리스

R2021a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by