86e8f7ab32cfd12577bc2619bc635690-AuthorFeedback.pdf

InTable3,16 different categories having different variants that perform well is likely due to specific shape distribution for each17 category. Similar toTable2,weobservethatmore complexshapes likeairplanes orchairs benefit from more views18 and information aggregation at both the feature and NOCS level. Other Questions: We do not employ any thresholding for the set union operation nor do we filter the estimated43 NOXmaps. In all experiments, we trained one network for each category as is normal practice45 but we also trained joint networks for all categories and are happyto include these results.