MeanSR: Restoration Trajectory Learning for One-Step Perceptual Super-Resolution
2026-08-10 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors propose MeanSR, a new method that creates high-quality sharp images from blurry, low-resolution ones all in a single step, making it faster than previous techniques that used slow, repeated processes. Instead of relying on existing complex models, their approach learns how to smoothly transform a low-quality image into a high-quality one by predicting the average change needed over time. They also introduce new ways to better train this transformation process. Tests show that MeanSR produces clearer and more realistic images while being quicker and less computationally expensive than some recent methods.
super-resolutiondiffusion modelsone-step inferenceaverage velocity fieldperceptual qualityimage restorationtrajectory matchingtemporal samplinginference latencyFLOPs
Authors
Axi Niu, Jiawei Kou, Kang Zhang, Qingsen Yan, Jinqiu Sun, Yanning Zhang
Abstract
Diffusion-based super-resolution (SR) achieves strong perceptual quality but requires costly iterative denoising. Existing one-step distillation methods reduce inference time but depend on expensive pretrained teachers, whereas CTMSR avoids distillation through PF-ODE consistency training yet does not explicitly model the restoration dynamics from low-resolution (LR) inputs to high-resolution (HR) images. We propose MeanSR, a one-step perceptual SR method that learns an LR-conditioned average velocity field to directly capture the finite-time transition from degraded or noisy inputs to plausible HR outputs. We further reformulate distribution trajectory matching for average-velocity generation and introduce a Stage-Aware Temporal Sampling strategy to improve trajectory learning. Experiments on synthetic and real-world benchmarks show that MeanSR outperforms CTMSR on CLIPIQA, MUSIQ, and MANIQA while substantially reducing FLOPs and inference latency. MeanSR also reconstructs sharper structures and more realistic textures with fewer perceptual artifacts.