TL;DR
Experimental results on multiple benchmark datasets demonstrate that the UGCL framework exhibits favorable robustness and competitive performance under both noisy and noise-free correspondence conditions.
Text-Based Person Retrieval (TBPR) aims to locate a person in an image database based on a natural language description. While effective in theory, TBPR faces substantial challenges in real-world scenarios due to noisy correspondences—misaligned or weakly related image-text pairs—that significantly degrade retrieval performance. Existing methods often overemphasize hard negative mining, which inadvertently magnifies the impact of such noise. To address this issue, we propose Dynamic Uncertainty with Noisy Correspondences (DUNC), a novel framework that incorporates two key components: (1) Cross-modal Evidential Learning (CEL), which models bidirectional alignment uncertainty using a Dirichlet distribution to capture the confidence in image-text similarity, and (2) Dynamic Robust Loss (DRL), which adaptively selects and aggregates hard negative samples to reduce the influence of noisy instances and improve model robustness. Unlike conventional global-alignment approaches, DUNC exploits fine-grained local correspondences to enhance semantic alignment between modalities. By integrating uncertainty-aware modeling and adaptive contrastive supervision, our method is capable of effectively disentangling noisy from reliable training pairs. Extensive experiments conducted on three benchmark datasets—CUHK-PEDES, ICFG-PEDES, and RSTPReid—demonstrate that DUNC consistently achieves state-of-the-art performance and exhibits strong robustness across a wide range of noise conditions. Code is publicly available at https://github.com/ASL-forever/DUNC.
Unsupervised person re-identification (USL-ReID) typically relies on clustering to generate pseudo-labels, but significant cross-view appearance variations often cause images of the same identity to be split into different clusters. Training on such noisy pseudo-labels severely degrades the learned representations. Therefore, learning robust view-invariant features is paramount. Data augmentation provides a direct way to enhance invariance, yet its trade-offs in USL-ReID remain under-explored: weak augmentations usually preserve identity semantics but lack diversity, whereas strong augmentations provide richer appearance diversity at the cost of partially corrupting identity-consistent semantic cues. To address this challenge, we propose Invariant Representation learning with Progressive Prototype Refinement (IRPP), a unified framework that learns invariant and discriminative features from noisy pseudo-labels. IRPP consists of three synergistic components. First, an Augmented Dual-Contrastive Learning (ADCL) module performs dataset-level prototype-guided invariant learning by contrasting weakly and strongly augmented views against cluster-derived prototypes. Second, an Alignment and Uniformity Learning (AUL) module regularizes the mini-batch-level weak–strong feature geometry, leading to more stable feature distributions under data augmentation. Third, a Progressive Prototype Refinement (PPR) mechanism progressively optimizes cluster centroids into cleaner prototypes, thereby mitigating the influence of noisy pseudo-labels and further strengthening invariant representation learning. This closed-loop design enables prototype-guided contrastive learning, weak–strong regularization, and prototype refinement to mutually reinforce each other. Extensive experiments on standard USL-ReID benchmarks demonstrate that IRPP achieves state-of-the-art performance with a simple and efficient training pipeline. Code is available at https://github.com/Trangle12/IRPP
Structural-Semantic Reciprocal Learning (SSRL), a framework that transforms open-loop association into a self-correcting closed-loop system, achieves robust cross-modal representation through the reciprocal interaction between structural and semantic learning.
Text-based person search (TPS) suffers from cross-modal informational skewness: pedestrian images are high-dimensional and redundancy-prone, while textual descriptions are sparse, incomplete, and sometimes inaccurate. To address the low alignment accuracy and poor robustness caused by the inherent uneven information distribution of visual and textual modalities in TPS, this paper proposes a unified Partition-based Information Rebalancing (PaIR) framework to realize balanced optimization and precise alignment of cross-modal information from both global content and local part dimensions. The framework adopts the CLIP dual-modal encoder for basic feature extraction and constructs a parallel global–local dual representation system to compensate for the lack of fine-grained spatial information in single global features. To eliminate modal redundancy and noise interference, a dual-modal noise suppression module is designed to filter invalid redundant information through visual foreground–background separation and textual token weight screening, while introducing adversarial constraints and orthogonal constraints to purify effective features. On this basis, a part balance alignment module is built to complete human semantic part decomposition and soft matching alignment for dual-modal features. Aiming at the common part semantic missing problem in textual descriptions, a visual part correlation affinity matrix is utilized for semantic associative completion to balance the information density of dual modalities. Finally, a global–local joint alignment strategy integrates hierarchical features and bidirectional cross-modal attention interaction to eliminate global–local semantic discontinuity and enhance fine-grained cross-modal matching capability. Extensive experiments on three public benchmarks demonstrate that PaIR consistently improves multiple baselines.
We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.
