IMPROVED PEDESTRIAN RE-IDENTIFICATION MODEL WITH ATTENTION MECHANISM AND MULTIPLE LOSS
Pedestrian Re-identification(RE-ID)concentrates on identifying specific pedestrians in cross-domain cameras.At present,most RE-ID algorithms study methods that enhance feature extraction ability,and various models appear,but they all have problems such as high model complexity or weak recognition ability.To solve these problems,BagTricks,a concise Re-ID benchmark model,was combined with channel attention mechanism to improve the ability of the model to extract significant features.Meanwhile,Circle loss was added to improve the loss function.Experiments on three popular Re-ID dataset show that this model obtains rank-1 accuracy of 95.6％and mAP of 88.5％on the mainstream image re-identification Market1501 dataset,and achieves the rank-1 accuracy of 89.1％and 76.7％on DukeMTMC and CUKE03 dataset.This method improves the accuracy of the model,and is easy to implement,and achieves competitive performance,which is better than most existing methods.
Pedestrian re-identificationDeep neural networkAttention mechanismMultiple loss function
万方数据
维普
