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

Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection #304

Open
chullhwan-song opened this issue Feb 17, 2020 · 1 comment

Comments

@chullhwan-song
Copy link
Owner

https://arxiv.org/abs/1912.02424
https://github.com/sfzhang15/ATSS

@chullhwan-song
Copy link
Owner Author

chullhwan-song commented Apr 29, 2020

Abstract

  • FCOS의 기반한 연구임.
    • 공개소스도 이 소스 base로 시작.
    • 그래서 실험을 보면, FOS+ATSS결합된 형태임.
  • 연구의 핵심은 anchor를 어떻게 sampling 할것인가에 대한 논의 - "how to define positive and negative training samples" > Adaptive Training Sample Selection (ATSS)

Adaptive Training Sample Selection

  • how to define positive and negative training samples
  • sampling 전략이 핵심인듯..
    • focal loss와 유사하나, 샘플링 전략에서 anchor와 ground-truth box와 center 와 가장 가까운 anchor를 찾고 여기에서, focal loss는 hyper-parameter 로써, IOU threshold를 주는 것같은데, 여기서는 이를 가까운 anchor와의 sampling 하여 이를 수치하하여 적용하는 것같다.
      • 참고) 소스보면 논문의 Algorithm 1에서의 단순함과는 차이 있다. ㅠ

@chullhwan-song chullhwan-song reopened this Jun 2, 2020
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