🤖 AI Summary
Existing text-to-video retrieval methods struggle to distinguish salient visual content from background noise under sparse textual queries, leading to inefficient cross-modal feature interaction. Inspired by human visual cognition, this work proposes a coarse-to-fine alignment framework that introduces, for the first time, both macro- and micro-perception mechanisms into video representation learning. Specifically, a Frame-Level Filtering Module (FFSM) eliminates temporal redundancy, while a Patch-Level Focusing and Compression Module (PFCM) aggregates salient visual entities, collectively emulating human-like visual attention. The proposed approach achieves state-of-the-art performance across five benchmark datasets, significantly improving both retrieval accuracy and semantic alignment between text and video modalities.
📝 Abstract
The success of CLIP has driven substantial progress in text-video retrieval. However, current methods often suffer from"blind"feature interaction, where the model struggles to discern key visual information from background noise due to the sparsity of textual queries. To bridge this gap, we draw inspiration from human cognitive behavior and propose the Human Vision-Driven (HVD) model. Our framework establishes a coarse-to-fine alignment mechanism comprising two key components: the Frame Features Selection Module (FFSM) and the Patch Features Compression Module (PFCM). FFSM mimics the human macro-perception ability by selecting key frames to eliminate temporal redundancy. Subsequently, PFCM simulates micro-perception by aggregating patch features into salient visual entities through an advanced attention mechanism, enabling precise entity-level matching. Extensive experiments on five benchmarks demonstrate that HVD not only captures human-like visual focus but also achieves state-of-the-art performance.