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

[4주차] VQ-VAE 코드 분석 #11

Open
HwangJohn opened this issue May 7, 2022 · 2 comments
Open

[4주차] VQ-VAE 코드 분석 #11

HwangJohn opened this issue May 7, 2022 · 2 comments
Assignees

Comments

@HwangJohn HwangJohn changed the title [3주차] VQ-VAE 코드 분석 [4주차] VQ-VAE 코드 분석 May 7, 2022
@HwangJohn
Copy link
Collaborator Author

HwangJohn commented May 7, 2022

  • vavae_ema 모드 에서 q_latent_loss 찾기

@HwangJohn
Copy link
Collaborator Author

HwangJohn commented May 8, 2022

  • vavae_ema 모드 에서 q_latent_loss 찾기

Because this loss term is only used for updating the dictionary, one can alternatively also update the dictionary items as function of moving averages of (not used for the experiments in this work). For more details see Appendix A.1.

논문의 3.2 Learning 수식 (3) 바로 위 단락에 있는 위 문구에 답이 있는거 같아요.
수식 (3)의 두번째 텀은 EMA로 대체될 수 있다라고 이해했고, 두번째 텀 목적이 codebook 임베딩 벡터 를 인코더 아웃풋들 의 중심과 가깝도록 업데이트 하는 것인데, K-means를 사용하고 싶으나(Appendix. 1), 미니배치 상에서는 전체 데이터에서의 중심점을 알 수 없으니 EMA를 통해서 각 배치마다 의 MA에 가깝도록 를 학습시키게 했네요.

@jwkanggist
위 내용 확인 부탁드려요 :)

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