awesome-davian / awesome-reviews-kaist

Computer vision paper reviews written by KAIST AI students
Creative Commons Attribution Share Alike 4.0 International
43 stars 117 forks source link

[2022 spring] ICLR 2022 VOS: LEARNING WHAT YOU DON’T KNOW BY VIRTUAL OUTLIER SYNTHESIS (20227059) #474

Open MonumentalCloud opened 2 years ago

MonumentalCloud commented 2 years ago

Thank you for such a detailed and insightful review of a field rarely in the spotlight. I've definitely learned a lot from reading your review and look forward to applying this knowledge to my own practice. While overall article was easy to read and impactful, I had a few nitpicking to talk about.

  1. For the beginners of machine learning, it might be good to give a more general explanation of what Out of distribution means.
  2. It would be better if the languages used are more beginner friendly. Ergo, it would be good to have a more general explanation in layman's terms why this matters, without the language used in the paper itself.

Thank you for such a detailed review!

Junhyeon-Park commented 2 years ago

좋은 리뷰 감사합니다. Out-of-Distribution(OOD) 기존 방법들을 소개하고, 이를 더 효율적으로 수행 가능한 virtual outlier에 대한 아이디어 및 방법을 매우 잘 정리해주셨습니다. Experiment에서는 데이터를 모두 정리하고 결과는 매우 잘 정리되었습니다. 하지만 3.1. 부터 Virtual Outlier Synthesis 설명해주실 때, 수식을 Problem definition 와 똑같이 자세히 설명하면 더 좋을 것 같습니다.

drumpt commented 2 years ago

Object Detection의 OOD를 해결하기 위해 feature space 상에서의 virtual outlier를 생성하는 VOS에 대한 전반적인 내용을 이해할 수 있었습니다. 전반적으로 잘 기술된 같습니다. 좋은 리뷰 감사드립니다. 몇 가지 피드백이 있는데요,