Person re-identification (re-id) plays a vital role in surveillance and forensics application. Since the labeled images for person re-id task is limited, the generalization ability of existed person re-id models is poor. On the other hand, images of different classes (pedestrian and non-pedestrian images) share some general features. To this end, this paper aims to improve the performance of person re-id by designing a relearning network which can learn domain-specific features and general features simultaneously. The proposed relearning network consists of a pretrained backbone network which provides the general features, and several attention-based subnetworks that learn domain-specific features from general features of different levels. Besides, we propose a coarse-fine loss to improve the generalization of person re-id model by making full use of the massive labeled non-pedestrian images. Experimental results on the publicly available Market-1501, DukeMTMC-reID and CUHK03 pedestrian re-id datasets demonstrate the effectiveness of the proposed relearning network and coarse-fine loss.INDEX TERMS Attention, coarse-fine, domain-specific features, general features.
Similar information purification and mining methods in the past are generally of low precision and weak usability. Therefore, we propose a method to dynamically update time series, that is, a similar information high precision purification and mining method based on time series updating. The method is used to implement regional linear time similar information time series by using rise analysis and linear regression analysis. Extreme value standardisation method is used to collate linear region so that the data in time series can be compared in parallel and the description of similar information feature is realised. Vertically align the two head-ends of time series to be purified; high precision purification is achieved by calculating the similarity of characteristics similarity displacement representation between two segments of similar information. Experimental verification shows that compared with previous methods, the recall F * value is the highest among different methods for five dataset, and the time cost of the proposed method is shorter than other methods. It was believed that .purification and mining performance of the proposed method is stronger with shorter time cost.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.