Rethinking data collection for person re-identification: active redundancy reduction. (May 2021)

• Record Type: Journal Article
• Authors: Xu, Xin; Liu, Lei; Zhang, Xiaolong; Guan, Weili; Hu, Ruimin
• Is Part Of: Pattern recognition. Volume 113(2021)
• Publication Date: 2021-05
• Subjects: Person re-identification -- Redundancy reduction -- Active learning; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.107827
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: To address the data collection problem in person reID, we present a novel active redundancy reduction framework to alleviate the data redundancy problem in public re-ID datasets. To minimize the annotation workload while maximizing the performance of the re-ID model, a simple baseline is presented to select informative and diverse samples for annotation by estimating their uncertainty and intra-diversity. A computer-assisted Identity Recommendation Module is proposed to help the human annotators to rapidly and accurately label the selected samples. Abstract: Annotating a large-scale image dataset is very tedious, yet necessary for training person re-identification (re-ID) models. To alleviate such a problem, we present an active redundancy reduction (ARR) framework via training an effective re-ID model with the least labeling efforts. The proposed ARR framework actively selects informative and diverse samples for annotation by estimating their uncertainty and intra-diversity, thus it can significantly reduce the annotation workload. Moreover, we propose a computer-assisted identity recommendation module embedded in the ARR framework to help human annotators to rapidly and accurately label the selected samples. Extensive experiments were carried out on several public re-ID datasets to demonstrate the existence of data redundancy. Experimental results indicate that our method can reduce 57%, 63%, and 49% annotation efforts on the Market1501, MSMT17, and CUHK03, respectively, while maximizing the performance of the re-ID model.