PHA-Net：基于原型的文本-视频检索分层对齐网络
原标题：PHA-Net: Prototype-based Hierarchical Alignment Network for Text-Video Retrieval
AI 摘要
本文提出了一种名为PHA-Net的原型分层对齐网络，用于文本-视频检索任务。该方法通过引入模态共享原型作为桥梁，并设计原型支持的令牌合并模块和原型对比损失，以高效优化跨模态对齐。在MSR-VTT、ActivityNet、VATEX和Charades四个基准上，PHA-Net在召回率总和上分别提升了8.8%、19.2%、0.7%和4.9%，验证了其有效性。
正文节选
Computer Science > Information Retrieval Title:PHA-Net: Prototype-based Hierarchical Alignment Network for Text-Video Retrieval View PDF HTML (experimental) Abstract:With the emergence of large-scale image-text pre-training models, e.g., CLIP, text-video retrieval has experienced substantial advances in recent years. Existing best-performing methods involve aligning cross-modal semantics at individual, local, and global levels simultaneously, raising concerns about the intrinsic sema