993edc98ca87f7e08494eec37fa836f7-AuthorFeedback.pdf

The reviewer notes that it would be great to include an analysis about how the pre-alignment of the objects in the6 ModelNet [35] and ShapeNet [5]dataset influences thequality ofthelearned representations. Inprevious literature,7 supervised tasks have consistently benefitted from pre-aligned objects by a small but significant margin (e.g. The reviewer wonders how much of the performance gain of our method with regards to previous25 unsupervised methods (i.e. FoldingNet) really stems from the improved architecture of DGCNN or the proposed26 task. The reviewer points out that some additional ablation study might be beneficial.