Pedestrian Attribute Recognition

Pedestrian attribute recognition (PAR) aims to automatically identify various characteristics of pedestrians in images or videos, such as gender, clothing, and carrying objects. Current research emphasizes improving accuracy and efficiency, focusing on transformer-based models and vision-language fusion approaches, often leveraging pre-trained models like CLIP, and addressing challenges like imbalanced datasets and attribute co-occurrence. These advancements are significant for applications in video surveillance, person re-identification, and other human-centric computer vision tasks, driving the development of more robust and generalizable systems.