The paper present some clear novelty in the under-studied domain of point-wise dependency estimation, with clear supporting evidence though various experiments. The reviewer concerns were clarified in the rebuttal, but we expect the authors to work significantly by including the changes they mention in their rebuttal, e.g: - shorten Section 3.1 and include more motivations for our presented approaches in Section 3.2. - include the MI discussion, - include ImageNet results in Experiment 2