NeurIPS 2020

ICNet: Intra-saliency Correlation Network for Co-Saliency Detection

Meta Review

The initial ratings were 5666. The main concerns were: 1) incremental novelty; 2) missing ablation on other datasets; 3) whether "Tree" group in MSRC experiments was also removed for other methods; 4) only small gains on some datasets In the response, authors clarified the contribution. They also provide additional ablation studies and provide reasons for the small gains on MSRC and iCoseg since they usually only contain one category of salient objects, whereas CoSal2015 has much larger proportion of multi-category saliency objects and the proposed method performs significantly better. After the response and discussion phase, all reviewers kept their original ratings. In particular, R2 was satisfied with most responses, except for the question about the "Tree" group being removed, which was not answered directly. The AC emailed the authors to clarify the point and they responded "Yes, for fair comparisons, in this paper, all methods (ours and the others) are evaluated on the MSRC without the "Tree" group.". Thus, the AC thinks this concern is addressed. Since all concerns were addressed, the AC recommends accept. Authors should update the paper according to the reviews and responses.