Text Based Person Retrieval

Text-based person retrieval (TPR) focuses on identifying individuals in images based solely on textual descriptions, bridging the gap between visual and linguistic information. Current research emphasizes improving model architectures by incorporating bidirectional embeddings, leveraging large language models for data augmentation, and refining alignment techniques to capture both positive and negative attributes within descriptions, often using contrastive learning methods. These advancements aim to enhance retrieval accuracy and robustness, particularly addressing challenges like limited training data and the inherent differences between visual and textual representations. Improved TPR has significant implications for various applications, including security, law enforcement, and multimedia search.