Feature semantic alignment and information supplement for Text-based person search

The goal of person text-image matching is to retrieve images specific pedestrians using natural language. Although a lot research results have been achieved in persona matching, existing methods still face two challenges. First,due the ambiguous semantic information features, aligning textual features with their corresponding image always tricky. Second, absence each local feature poses significant challenge network extracting robust that match both modalities. To address these issues, we propose model for explicit extraction and effective supplement. On one hand, by attaching consistent clear information, course-grained alignment between achieved. other an supplement proposed, which captures relationships modality supplements them obtain more complete information. In end, are then concatenated comprehensive global feature, capable precise described features. We did extensive experiments on CUHK-PEDES dataset RSTPReid dataset, experimental show our method has better performance. Additionally, ablation experiment also proved effectiveness module designed this paper.