Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fully Convolutional Networks for Semantic Segmentation #74

Open
chullhwan-song opened this issue Jan 31, 2019 · 1 comment
Open

Fully Convolutional Networks for Semantic Segmentation #74

chullhwan-song opened this issue Jan 31, 2019 · 1 comment

Comments

@chullhwan-song
Copy link
Owner

chullhwan-song commented Jan 31, 2019

https://arxiv.org/abs/1411.4038

@chullhwan-song
Copy link
Owner Author

chullhwan-song commented Jan 31, 2019

Abstract

  • key insight - “fully convolutional” = FCN 논문의 최초 논문이라할수 있음.
  • segmentation task
    • alexnet, vgg, resnet등에서 이를 이용하여 segmentation task를 푸는게 가능
      image
    • 사실 이 그림 하나로 설명이 다 됨
    • 따라서, base 구현은 그닦 어렵지 않을듯~

Fully convolutional networks

  • 각 convnet는 feature map 즉, c x w x h 차원
    • spatial information = wxh
    • feature = c
  • Convnets are built on translation invariance
    • Their basic components (convolution, pooling, and activation functions) operate on local input regions, and depend only on relative spatial coordinates
  • 기존 이미지분류에 쓰이는 alexnet/vgg등에서 맨 뒤에 존재하는 fully connected layer를 제거하고 convolution layer로 대체 (1x1)하는 형태 = convolutionalization
    image
    • object에 대한 위치 정보를 보존된다.다음그림이 인터넷에..더 좋음.
      image
    • 실제 이를 원본크기로 re-scale하면됨
      • bi-linear interpolation 가능하지만, 학습상에서는 deconvolution을 이용한듯.. 하지만 upsampling할때 최근에 bi-linear interpolation도 많이 이용하는듯~
  • 실제로 skip-connection을 이용하여 더 성능 향상
    image

실험

image
image

결론

  • FCN 이후에 매우 강력한 알고리즘이란걸 알게됨 - 난 약식으로 리뷰하지만,
  • 인용지수만보더래도..
  • 대부분의 segmentation, localization에서 여향을 미쳤다고 봄(개인 의견)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

1 participant