A Spatio-Temporal Hierarchical Diffusion Framework for Training-Free Perceptual Remote Sensing Image Compression (opens in new tab)

Remote sensing (RS) data contains vast amounts of geographic structural information and fine-grained textural details, making compression essential and necessitating efficient processing for downstream interpretation tasks. Generative models have shown great potential in learned image compression (LIC), offering high compression rates and exceptional generation quality. Among these, diffusion models stand out for their ability to produce high-quality reconstructions. However, they may halluci...