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

[7] Self Supervision for Attention Networks #7

Open
techzzt opened this issue Jul 13, 2022 · 0 comments
Open

[7] Self Supervision for Attention Networks #7

techzzt opened this issue Jul 13, 2022 · 0 comments
Labels
CVPR Computer Vision and Pattern Recognition

Comments

@techzzt
Copy link
Owner

techzzt commented Jul 13, 2022

Self Supervision for Attention Networks

Patro, Badri N., et al. "Self supervision for attention networks." Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 2021.

기존의 attention mechanism의 한계를 언급하고 self-supervision을 도입해 attention mechanism의 성능적인 향상을 증명함

image
Figure 2. Generation of Surrogate Attention Maps for visualattention based network.

  • Attention module을 이용해 attention map을 추출했으며 해당 방법론은 어떤 모델을 사용하더라도 적용 가능함
  • 방법론의 입력 데이터는 image, question (text modality) 형태이며 각 형태에 맞는 모델 (CNN, LSTM)의 output으로 attention을 계산함
  • 모델은 입력 데이터의 각기 다른 영역에서 확률 값(attention probability)을 추출해 사용하며 해당 확률 값으로 계산된 attention distribution은 supervision을 위해 사용됨
  • Supervision을 기반으로 모델은 salient한 영역에 더욱 집중할 수 있으며 해당 부분을 통해 accuracy 향상이 가능해짐

image

  • 제안된 attention map과 masking을 제외한 attention (original)을 출력한 결과, surrogate attention이 특징 적인 영역을 좀 더 세밀하게 attention을 부여하고 있음을 확인 가능함
@techzzt techzzt added the CVPR Computer Vision and Pattern Recognition label Jul 13, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
CVPR Computer Vision and Pattern Recognition
Projects
None yet
Development

No branches or pull requests

1 participant