NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:7280
Title:A Fourier Perspective on Model Robustness in Computer Vision


		
The paper proposes an interesting angle to investigate robustness of convnets, looking at Fourier spectrum of images and/or perturbations. Two of the reviewers were very positive, while R2 raised some concerns and ultimately was not completely satisfied by the rebuttal. However, in discussion among the reviewers and the AC, everybody agreed that the paper has potentially important contributions. Despite the shortcomings I recommend to accept the paper as a poster. I do recommend that the authors aim to take the detailed comments and improvement suggestions by all the reviewers into account. [This meta-review was reviewed and revised by the Program Chairs]