Pyramid-resolution person restoration for cross-resolution person re-identification

Abstract

In this study, we present a simple yet effective pyramid-resolution person restoration method for cross-resolution person re-identification. Our method involves a pyramid resolution restoration network that enhances pyramid resolution images, and utilizes feature distance fusion to leverage valuable and complementary information from these pyramid images. Extensive experiments demonstrate the effectiveness of our method on both real-world cross-resolution datasets and simulated datasets.