ReSAGE-PAR: Representational Similarity Assessment for Generative Expansion in Pedestrian Attribute Recognition
Researchers have developed ReSAGE-PAR, a novel pipeline designed to enhance pedestrian attribute recognition (PAR) by leveraging diffusion models for image synthesis. This method addresses challenges like domain gaps between training data and surveillance images, and the risk of generative hallucinations. ReSAGE-PAR adapts diffusion models to target resolutions and uses vision-language alignment scores to generate reliable pseudo-labels, significantly improving PAR performance. AI
IMPACT Enhances pedestrian attribute recognition by enabling scalable, high-fidelity dataset expansion using generative models.