NeurIPS 2020

Hierarchical Quantized Autoencoders


Meta Review

Post-rebuttal, 3 out of 4 reviewers vote for acceptance while R1 esteems it is marginally below threshold. All 4 reviewers appreciated the convincing results in low-bitcount regimes. The two main points debated in the discussion phase concerned: a) whether there was sufficient novelty in the work w.r.t. VQ-VAE related prior-works. For this, R4 convincingly argued that the paper significantly contributes to the learned compression sub-field. W.r.t. VQ-VAE, taking into account the author response and following reviewer discussion, the AC agrees that there are original aspects in the proposed algorithm that make it markedly different, with reasonable justification and ablation. The AC judges that the new outlook proposed will be useful for the research community to ponder. b) whether the rather small scale experiments (in particular the 64x64 resolution) are sufficient by current standards. -> for this authors said in their reply that they would add results of experiments with higher resolution. A remaining criticism of R1 is that the ideas and model should be presented more clearly. Given its judged valuable contribution to the learned compression sub-field, and the original aspects of the approach being worth pondering by the community interested in learning quantized representations, the AC recommends acceptance. But in accordance with the reviewes requests that the final revision: (a) clarifies section 4 and 5. (b) includes experiments at higher resolution as promised by the authors.