Towards Robust Text-to-Image Person Retrieval: Multi-View Reformulation for Semantic Compensation

TL;DR

This paper introduces a multi-view semantic reformulation framework driven by LLMs to improve robustness in text-to-image person retrieval by addressing expression drift and enhancing cross-modal consistency.

Contribution

It proposes a novel multi-view reformulation and feature compensation approach that boosts retrieval accuracy without additional training.

Findings

Achieves state-of-the-art results on three datasets.

Enhances cross-modal consistency through multi-view semantic reformulation.

Improves robustness without additional training.

Abstract

In text-to-image person retrieval tasks, the diversity of natural language expressions and the implicitness of visual semantics often lead to the problem of Expression Drift, where semantically equivalent texts exhibit significant feature discrepancies in the embedding space due to phrasing variations, thereby degrading the robustness of image-text alignment. This paper proposes a semantic compensation framework (MVR) driven by Large Language Models (LLMs), which enhances cross-modal representation consistency through multi-view semantic reformulation and feature compensation. The core methodology comprises three components: Multi-View Reformulation (MVR): A dual-branch prompting strategy combines key feature guidance (extracting visually critical components via feature similarity) and diversity-aware rewriting to generate semantically equivalent yet distributionally diverse textual variants; Textual Feature Robustness Enhancement: A training-free latent space compensation mechanism suppresses noise interference through multi-view feature mean-pooling and residual connections, effectively capturing "Semantic Echoes"; Visual Semantic Compensation: VLM generates multi-perspective image descriptions, which are further enhanced through shared text reformulation to address visual semantic gaps. Experiments demonstrate that our method can improve the accuracy of the original model well without training and performs SOTA on three text-to-image person retrieval datasets.