Person Re-identification Method Based on Dual Feature Attention Backbone Network

Person re-identification using attention mechanisms has been a prominent research focus, enabling the extraction of highly discriminative features.. Nonetheless, current pedestrian re-identification algorithm networks exhibit deficiencies in utilizing partially occluded image features, extracting fine-grained details inadequately, and capzturing features at lower levels.. To tackle these challenges, we introduce a person re-identification algorithm leveraging the Dual-Feature Attention Backbone Network (DFABnet). Initially, a pixel-wise attention module is incorporated to enhance the receptive field and extract features at a higher level. Subsequently, a feature completion module is implemented to address the incomplete parts of obscured features. Lastly, a fine-grained channel attention module is devised to extract finely detailed features. Experimental results on the Market 1501 and DukeMTMC-reID datasets validate the efficacy of the proposed model.