The paper presents a novel approach to predicting 3D shape of objects from image(s). The reviews were split, with R1 being rather positive, R3 quite negative and R2 marginally positive. Following the rebuttal and discussion period, this score discrepancy remained. I have read the paper, the reviews and the rebuttal, and despite some concerns raised by R3 (and in part agreed with by R1) I side with the (weak) majority vote, and feel that the paper has enough novelty and is likely to generate enough interest to be accepted. The proposed approach builds upon the previously proposed methodology (NOCS) but extends it in a substantial way, incorporating reasoning about the "back-facing" surface points of the object. It also proposes a novel way to aggregate (variable number of) multiple views for better reconstruction. The authors can improve the paper significantly just by incorporating into the final version various responses they included in the rebuttal.