NeurIPS 2020

The Dilemma of TriHard Loss and an Element-Weighted TriHard Loss for Person Re-Identification

Meta Review

All four knowledgeable reviewers were left with a favorable opinion of this work after the author rebuttal, and the AC agrees with this positive assessment. However, during the post-rebuttal discussion phase an independent ethics review was conducted regarding the general use of the DukeMTMC dataset. Some concerns raised about use of this dataset during this ethics review include: -- "... the dataset collection involved non-consensual video surveillance of students on Duke University campus. It is unlikely that all students even knew they were being recorded, and their relative lack of power with respect to the institution surveilling them also raises concerns about the ability to meaningfully object to the surveillance." -- "Including the dataset in the paper as-is would be problematic, as it would contribute to this mainstream use of the dataset. Referencing the issues and discouraging future use of the dataset would help mitigate this, as would full removal of the results." -- "The fact that others use the dataset uncritically does not make it appropriate, but it could have contributed to the authors being unaware of the issue, and that awareness may vary geographically." The recommendation of the NeurIPS Ethics Panel is that that *the DukeMTMC dataset should NOT be used in an accepted NeurIPS paper*. The AC therefore recommends a **conditional accept** (a rare and exceptional case reserved only for papers with ethical issues). This means the paper is accepted on the condition that the final version 1) removes all results on DukeMTMC, 2) makes it clear that the dataset has been taken down and should no longer be used, and 3) reproduces all technical contributions and results of the original submission using other datasets. Moreover, the AC strongly recommends that the authors use the Broader Impact section to state clearly that surveillance is the typical goal of re-id systems, that re-id systems often rely on non-consensual surveillance data for their training, and to discuss and generally raise awareness about the consent challenges with MTMC data collection. The authors should also address the suggestions of R3 regarding better evidence in support of key claims and improvements to the writing. The AC discussed this decision with the SAC and PCs. ******************************* Note from Program Chairs: The camera-ready version of this paper has been reviewed with regard to the conditions listed above, and this paper is now fully accepted for publication.