What are the contributions of the paper: This paper presents an automatic video based ReID system. Occlusion Filter and False Positives Class are used to remove the interference of unreliable detections. The ReID task is extended from individual images to video clips through a proposed Window-based Classifier. And at last, new metrics are presented to evaluate the whole system.
What are the additional ways in which the paper could be improved: I think the problem is important and the proposed sulution is reasonable, with some incremental novelties. The clip-based output is a good display mode in the application. However, I have several questions and list them below.
1. The references are not enough. Most recent related works, such as ReID papers in CVPR2014,15, are not included.
Hmm, indeed. We thank you for calling our attention to this. We have added some references and now have four 2014 papers, two of those from CVPR, and seven 2015 CVPR papers.
2. In Chapter 2, the organization could be improved. The two main parts are the taxonomy and the overview, which should be distinguished clearly or split into two subsections. What’s more, there’s no analysis after the taxonomy. And what’s this taxonomy used for?
Agreed, we have now divided Chapter 2 into a Taxonomy and an Overview sections. We have also added a paragraph before the Overview section as follows:
This taxonomy allows for a better understanding of the dimensions of the RE-ID problem, and thus be able to better organize an analysis of the state-of-the-art in it’s several categories.