NeurIPS 2020

Detecting Hands and Recognizing Physical Contact in the Wild

Meta Review

The initial scores for this paper were: 4: An okay submission, but not good enough; a reject. 6: Marginally above the acceptance threshold. 6: Marginally above the acceptance threshold. This is a borderline case. The positive points are: - The work addresses an important and less explored problem of detecting physical contact states. - The work provides sound empirical evaluation with a range of ablation studies, comparison to prior work and well-chosen baselines. - New dataset with good quality control that will be useful for the community. The negative points: - Relatively small improvements over the state-of-the-art (4%), though the improvements are consistent. - Missing experiments. - Some clarity issues. - Better description of relation to prior work is needed. - Limited technical novelty. - Relatively small size of the dataset (18k training images, 1.6k test images). The authors provide a rebuttal, which addresses many of the weak points. In the post-rebuttal discussion, the negative reviewer R1 (4) upgrades their score from 4 to 6. R3 confirms their positive rating. The final scores are 6, 6, 6 with all reviewers leaning towards acceptance. The AC is convinced by the positive arguments of the reviewers and recommends Accept. The authors are strongly encouraged to take into account all reviewers' feedback when preparing the final version.