NeurIPS 2020

Compressing Images by Encoding Their Latent Representations with Relative Entropy Coding

Meta Review

The approach is novel, presented in a nice and clear way and with good results. After the rebuttal and discussion phase, there was a generalized consensus that this submission should be accepted to the conference. That said, a reviewer raised concerns regarding an ablation experiment on the effect of importance sampling. I would encourage the authors to consider, space permitting, such an addition, or at least to add a sentence to clarify this. Some typos are also present and should be fixed.