Reviewers felt that this is a strong submission with a relatively simple idea that produces a substantial improvement over baseline methods. The paper is fairly well-written. The experimental results are comprehensive, and reviewers especially appreciated the experiments showing the impact of compression on downstream tasks. Reviewers felt that the proposed method is clever and interesting. There was some concern about the technical novelty in relation to prior work (e.g. OctSqueeze), as well as some missing baseline comparisons. There were also questions about the number of forward passes, runtime, GPU memory, etc.