Weakly Supervised Person Search

Weakly supervised person search tackles the challenge of simultaneously detecting and identifying individuals in images using only bounding box annotations, eliminating the need for laborious identity labels. Recent research focuses on improving feature learning through techniques like contrastive learning (particularly intra-image approaches addressing spatial and occlusion variations) and leveraging self-similarity across different image scales. These advancements aim to enhance the accuracy and robustness of person search models, potentially impacting applications like video surveillance and visual search by reducing the reliance on extensive manual annotation. The incorporation of contextual information and strategies to handle unpaired persons further improves performance.