SPrHrNet: Style-Preference Hierarchical Net for Fashion Retrieval with Text Feedback

Existing algorithms in fashion retrieval with text feedback face challenges such as insufficient attention to user needs, inability to simultaneously focus on global and local features of clothing, and unclear structure between features and categories. To address these issues, we propose Style-Preference Hierarchical Net (SPrHrNet), which emphasizes clothing style and user preferences within a hierarchical structure, achieving better fine-grained text-image alignment. Our model includes a Style-Preference Controller (SPrCr) module to simultaneously emphasize global and local features, and a Hierarchical Encoder (HE) module to preserve fine-grained structure among features and categories. Experiments demonstrate the effectiveness and superiority of SPrHrNet compared to other advanced fashion retrieval algorithms.