Training-Free Rate-Distortion-Perception Traversal With Diffusion

Yuhan Wang, Suzhi Bi, Ying-Jun Angela Zhang

Keywords: Rate-Distortion-Perception Tradeoff, Training-Free, Diffusion Model, Lossy Compression

Abstract: The rate-distortion-perception (RDP) tradeoff captures the fundamental limits of lossy compression by jointly considering bitrate, reconstruction fidelity, and perceptual quality. While recent neural compression methods have improved perceptual performance, they typically operate at a fixed point on the RDP surface, requiring retraining to target different tradeoffs. In this work, we propose a training-free framework for traversing the full RDP surface, utilizing pretrained diffusion models. Our approach integrates a reverse channel coding (RCC) encoder with a novel score-scaled probability flow ODE decoder. We theoretically prove that the proposed decoder is optimal for the distortion-perception tradeoff under AWGN observations and that the overall framework with the RCC encoder is optimal for the RDP function in the Gaussian case. Empirical results across multiple datasets demonstrate the framework's flexibility and effectiveness in navigating the ternary RDP tradeoff using pre-trained diffusion models. Our results establish a practical and theoretically grounded approach to adaptive, perception-aware compression.

Primary Area: generative models

Submission Number: 10192