FreqOrtho-SR: Frequency-Guided Orthogonal Expert Learning for Real-World Image Super-Resolution

2026-06-27

AI Summary

This work addresses the challenge in real-world image super-resolution of simultaneously achieving high pixel fidelity and strong semantic perceptual quality, as well as the limited generalization of single-adapter approaches. To this end, we propose a frequency-guided Mixture-of-LoRA-Experts mechanism combined with an orthogonal gradient projection strategy. Specifically, frequency-domain features extracted via FFT enable degradation-aware dynamic expert routing, while singular value decomposition (SVD)-based subspace disentanglement and orthogonal gradient projection enforce decoupling between fidelity and perceptual objectives during optimization, thereby facilitating complementary learning. The proposed method achieves state-of-the-art performance in balancing fidelity and perceptual quality across multiple benchmarks and supports efficient single-step inference.

Abstract

Diffusion prior-based methods have shown impressive results in real-world image super-resolution (ISR), yet two key challenges persist: balancing pixel-level fidelity with semantic quality, and adapting to diverse degradations. Existing dual-branch approaches freeze the pixel module during semantic training, but the semantic branch can still expand capacity within the pixel subspace, precluding genuine perceptual improvement. Moreover, using a single static adapter cannot generalize across heterogeneous real-world corruptions. To address both issues, we propose FreqOrtho-SR, which comprises: $\textbf{Freq}$uency-guided Mixture of LoRA Experts (FreqMoE), it routes inputs to specialized experts via a non-parametric FFT-based degradation-feature extractor that encodes frequency-domain signatures, enabling stable and interpretable specialization across corruption types; and $\textbf{Ortho}$gonal Gradient Projection (OGP), which reframes the dual-objective optimization as a subspace-constrained problem: by extracting the pixel-fidelity subspace via SVD on combined expert weight deltas and projecting semantic gradients onto its null space, OGP guarantees orthogonality between the two objectives, enabling genuinely complementary learning without mutual interference. Experiments show that FreqOrtho-SR achieves competitive overall performance and a strong fidelity-perception trade-off across multiple benchmarks with efficient single-step inference. The source code of our method can be found at $\href{https://github.com/sonhm3029/FreqOrtho-SR}{\texttt{sonhm3029/FreqOrtho-SR}}$.

Problem

• image super-resolution
• fidelity-perception trade-off
• real-world degradations
• dual-branch optimization
• diffusion prior

Innovation

• Frequency-Guided Mixture of Experts
• Orthogonal Gradient Projection
• Diffusion Prior
• Real-World Super-Resolution
• LoRA