Generative Compression

Generative compression leverages deep generative models, such as VAEs, GANs, and diffusion models, to achieve efficient data compression while maintaining or enhancing perceptual quality. Current research focuses on improving rate-distortion-perception trade-offs, developing methods for controllable bitrate adaptation, and exploring the use of semantic guidance (e.g., through maps or segmentation) to improve reconstruction fidelity, particularly at extremely low bitrates. This approach offers significant potential for improving the efficiency of data storage and transmission across various domains, including remote sensing imagery, video coding, and on-device machine learning, by reducing computational costs and power consumption.