Person re-identification (re-id) is a critical problem in video analyticsapplications such as security and surveillance. The public release of severaldatasets and code for vision algorithms has facilitated rapid progress in thisarea over the last few years. However, directly comparing re-id algorithmsreported in the literature has become difficult since a wide variety offeatures, experimental protocols, and evaluation metrics are employed. In orderto address this need, we present an extensive review and performance evaluationof single- and multi-shot re-id algorithms. The experimental protocolincorporates the most recent advances in both feature extraction and metriclearning. To ensure a fair comparison, all of the approaches were implementedusing a unified code library that includes 11 feature extraction algorithms and22 metric learning and ranking techniques. All approaches were evaluated usinga new large-scale dataset that closely mimics a real-world problem setting, inaddition to 16 other publicly available datasets: VIPeR, GRID, CAVIAR,DukeMTMC4ReID, 3DPeS, PRID, V47, WARD, SAIVT-SoftBio, CUHK01, CHUK02, CUHK03,RAiD, iLIDSVID, HDA+ and Market1501. The evaluation codebase and results willbe made publicly available for community use.
展开▼