awesome-davian / awesome-reviews-kaist

Computer vision paper reviews written by KAIST AI students
Creative Commons Attribution Share Alike 4.0 International
41 stars 118 forks source link

[2022 spring] CVPR 2021 GIRAFFE: Representing Scenes as Compositional Generative Neural Feature Fields (20223110) #513

Open koo616 opened 2 years ago

koo616 commented 2 years ago

우선 좋은 리뷰 감사드립니다. Related work 섹션에서 리뷰해주신 논문과 관련된 이전 연구들을 이미지와 더불어 설명해주셔서 논문에 대한 이해도가 더 높아질 수 있었던 것 같습니다. 또한 Problem definition과 Idea 섹션에서 논문의 목표와 핵심 요소를 간단 명료하게 설명해주셔서 흐름을 파악하기 용이했습니다.

다만 Method 섹션에서 GIRAFFE 모델과 관련된 설명을 해주심에 있어서 수식은 명확하게 제시되었지만 이에 대한 부가적인 설명이 조금 더 있었다면 좋겠다는 생각이 들었습니다. 예를 들어, training 부분에 대해서 수식이 가진 간단한 의미가 추가된다면 모델에 대한 이해도가 더 높아질 것이라고 생각됩니다.

더불어 다소 작은 부분들이지만 Idea 섹션에서 Imagary라고 표현하신 부분이 imagery에서 오타가 난 것인 것 같아 확인해주시면 감사드릴 것 같고, Result 섹션에서 이미지가 더 크게 삽입된다면 더 쉽게 결과를 비교해볼 수 있을 것 같습니다.

Victor02091 commented 2 years ago

Interesting article, it could be usefull for my team project on style GAN.

-In part 3 after "Neural Feature Field" your LateX equation did not render (h_{\theta}:R^{L_x} \times R^{L_d} \times R^{M_s} \times R^{M_a} \to R^{+} \times R^{M_f}) -It could be interesting to explain a bit more the complex math formulas because sometimes it is hard to read and understand.