.png){kind=icon}
.png){kind=icon}
Return
OODDiffusion: A deep diffusion-based blind image super resolution scheme using out-of-distribution learning and controllable sampling process
.png){kind=icon}
PRE
AI
S
A
M
M
DOI：10.1016/j.imavis.2026.106076.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We adapt the decomposed confidence formulation for out-of-distribution detection to the BSR setting by incorporating a feature stability signal, enabling degradation-aware detection beyond conventional output-level confidence. • We propose an OOD-guided diffusion framework with explicit routing, where the detected distribution (ID vs. OOD) determines the sampling strategy for restoration. • We develop an adaptive noise scheduling strategy for OOD degradations based on a sigmoid-weighted schedule that emphasizes mid diffusion timesteps, improving robustness to degradation-distribution mismatch. • Extensive experiments demonstrate that the proposed method achieves strong performance compared to recent diffusion-based blind SR approaches, particularly on perceptual and no-reference quality metrics.
Journal
IF:
4.2
Papers:
4.0K
Citations:
6.7K
Organization
Cited Papers