Search papers, labs, and topics across Lattice.
This paper introduces UGDiff, a novel uncertainty-guided diffusion paradigm that enhances the perception-distortion trade-off in single-image super-resolution (SR). By estimating reconstruction uncertainty in latent features, UGDiff selectively restores high-frequency details in regions of high uncertainty while maintaining fidelity in other areas. Experimental results show that UGDiff significantly outperforms existing state-of-the-art diffusion-based SR methods, achieving a superior balance between perceptual quality and fidelity.
By intelligently guiding the diffusion process with uncertainty estimates, UGDiff achieves a groundbreaking balance between perceptual realism and fidelity in super-resolution tasks.
The perception-distortion trade-off poses a fundamental challenge in single-image super-resolution (SR). Although diffusion-based SR methods excel at generating perceptually realistic images, achieving high fidelity remains a key limitation. Recent advances in diffusion-based SR have shown promise in improving fidelity, but these methods often compromise perceptual quality due to their high reliance on a high-fidelity image. To address this, we introduce UGDiff, a novel diffusion guidance paradigm designed to further improve the perception-distortion balance. In particular, we first estimate the reconstruction uncertainty of the latent features corresponding to a high-fidelity image. This uncertainty is then used to guide the diffusion process to selectively restore high-frequency details in high-uncertainty regions, while preserving fidelity elsewhere. Furthermore, our guidance method adaptively identifies the high-uncertainty regions by considering not only the estimated uncertainty but also the posterior variance of the diffusion sampler at each timestep. This relaxes the reliance on the high-fidelity image in the later stages of sampling, thereby achieving a better perception-distortion balance. Extensive experimental results demonstrate that our method performs favorably against state-of-the-art diffusion-based SR methods.
