CTSR: Controllable Fidelity-Realness Trade-off Distillation for Real-World Image Super Resolution

2025-03-18

AI Summary

Real-world image super-resolution faces a fundamental trade-off between fidelity (measured by PSNR) and perceptual realism (measured by LPIPS). To address this, we propose a controllable fidelity–realness trade-off distillation framework. Our method introduces geometric decomposition to explicitly model the two objectives, enabling disentangled representation of fidelity and realness; designs a continuously adjustable interpolation mechanism for user-controllable balancing; and integrates multiple diffusion-based teacher models via dual-stream feature disentanglement and collaborative knowledge distillation to enhance generalization. Extensive experiments on multiple real-world super-resolution benchmarks demonstrate consistent superiority over state-of-the-art methods, achieving simultaneous and significant improvements in both PSNR and LPIPS—marking the first work to realize synergistic optimization of high fidelity and high perceptual realism.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Mixed Discrete/Continuous Search
• Machine Learning: Semi-Supervised Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Real-world image super-resolution is a critical image processing task, where two key evaluation criteria are the fidelity to the original image and the visual realness of the generated results. Although existing methods based on diffusion models excel in visual realness by leveraging strong priors, they often struggle to achieve an effective balance between fidelity and realness. In our preliminary experiments, we observe that a linear combination of multiple models outperforms individual models, motivating us to harness the strengths of different models for a more effective trade-off. Based on this insight, we propose a distillation-based approach that leverages the geometric decomposition of both fidelity and realness, alongside the performance advantages of multiple teacher models, to strike a more balanced trade-off. Furthermore, we explore the controllability of this trade-off, enabling a flexible and adjustable super-resolution process, which we call CTSR (Controllable Trade-off Super-Resolution). Experiments conducted on several real-world image super-resolution benchmarks demonstrate that our method surpasses existing state-of-the-art approaches, achieving superior performance across both fidelity and realness metrics.

Problem

• Balancing fidelity and realness in image super-resolution.
• Leveraging multiple models for improved super-resolution performance.
• Enabling controllable trade-off between fidelity and visual realness.

Innovation

• Distillation leverages multiple teacher models
• Geometric decomposition balances fidelity and realness
• Controllable trade-off enables adjustable super-resolution