Search papers, labs, and topics across Lattice.
The paper introduces CASR, a cyclic super-resolution framework designed to mitigate cross-scale distribution shift in arbitrary-scale super-resolution (ASISR). By reformulating ultra-magnification as a series of in-distribution scale transitions, CASR maintains stable inference across arbitrary scales using a single model. The framework incorporates a Structural Distribution Alignment Module (SDAM) to align structural distributions and a Self-similarity Aware Restoration Module (SARM) to restore high-frequency textures.
Achieve state-of-the-art arbitrary-scale super-resolution with a single model by reframing extreme upscaling as a series of distribution-aligned, self-similarity-aware transitions.
Arbitrary-Scale SR (ASISR) remains fundamentally limited by cross-scale distribution shift: once the inference scale leaves the training range, noise, blur, and artifacts accumulate sharply. We revisit this challenge from a cross-scale distribution transition perspective and propose CASR, a simple yet highly efficient cyclic SR framework that reformulates ultra-magnification as a sequence of in-distribution scale transitions. This design ensures stable inference at arbitrary scales while requiring only a single model. CASR tackles two major bottlenecks: distribution drift across iterations and patch-wise diffusion inconsistencies. The proposed SDAM module aligns structural distributions via superpixel aggregation, preventing error accumulation, while SARM module restores high-frequency textures by enforcing autocorrelation and embedding LR self-similarity priors. Despite using only a single model, our approach significantly reduces distribution drift, preserves long-range texture consistency, and achieves superior generalization even at extreme magnification.
