NeurIPS 2020

Fourier Spectrum Discrepancies in Deep Network Generated Images


Meta Review

The reviewers are in agreement that the observations made in this paper (fake images generated by GANs and VAEs are detectable from their high frequency statistics) are interesting and relevant to the NeurIPS community. After significant reviewer discussion, and after weighting the author response, the reviewer consensus is that the paper should be accepted. The reviewers did not identify any major shortcomings of the paper that were left unaddressed. I agree that the paper presents interesting results and I support acceptance.