Dynamic sparse and weight allocation-based text-driven person retrieval

doi:10.1016/j.imavis.2025.105737

Dynamic sparse and weight allocation-based text-driven person retrieval

2025-09-24

0

PRE

AI

DOI：10.1016/j.imavis.2025.105737

Original

Original request for help

Save

Abstract

En 中文

• We propose a progressive enhancement for a multimodal integration strategy that ensures that information at different levels of detail can be effectively integrated and aligned, thus improving the model’s ability to understand complex data. • We adopt a global two-way match filtering method, which can effectively mitigate the interference of false matching pairs of image text, so as to select the image text with a high matching degree for training. • We introduce a fine-grained dynamic sparse mask modeling approach, which not only extracts the key elements in the features but also mines the image text for fine-grained relationships, thus improving the retrieval performance.

Dynamic sparse and weight allocation-based text-driven person retrieval

Abstract

Journal

Organization

Cited Papers