NeurIPS 2020

GPS-Net: Graph-based Photometric Stereo Network


Meta Review

This paper was reviewed by four reviewers knowledgeable in photometric stereo. It received scores of 4,7,8, and 6. After reading the rebuttals and each other's reviews, none of the reviewers changed their opinion. There was a lot of discussion about this paper. R1 has some valid concerns about the ablation experiments. This has been partially resolved thanks to the additional experiments in the rebuttal, but R1 still had some remaining doubts. R2, R2, and R4 were more positive and felt that this work might open the door to further exploration of hybrid photometric approaches that combine per-pixel and all-pixel reasoning. They also commented on the qualitative and quantitative performance. Despite R1's reservations, this AC sides with the other reviews and thinks there is enough merit in accepting the paper. For the final paper, the authors are strongly encouraged to (i) add the ablation experiments from the rebuttal, (ii) improve the presentation of Fig 1 (see R1 and R2's comments), and (iii) add the ablation of Eq 5 promised to R1 and R2. Furthermore they should take the detailed comments of R1 into account, as they will further strengthen the paper.