RefReward-SR: LR-Conditioned Reward Modeling for Preference-Aligned Super-Resolution

• URL: http://arxiv.org/abs/2603.24198v1
• Date: Wed, 25 Mar 2026 11:20:51 GMT
• Title: RefReward-SR: LR-Conditioned Reward Modeling for Preference-Aligned Super-Resolution
• Abstract summary: RefReward-SR is a low-resolution (LR) reference-aware reward model for preference-aligned SR.<n>It assesses high-resolution (HR) reconstructions conditioned on LR inputs, treating the LR image as a semantic anchor.<n>Our framework achieves substantially better alignment with human judgments, producing reconstructions that preserve semantic consistency.
• Score: 26.50828843718706
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Recent advances in generative super-resolution (SR) have greatly improved visual realism, yet existing evaluation and optimization frameworks remain misaligned with human perception. Full-Reference and No-Reference metrics often fail to reflect perceptual preference, either penalizing semantically plausible details due to pixel misalignment or favoring visually sharp but inconsistent artifacts. Moreover, most SR methods rely on ground-truth (GT)-dependent distribution matching, which does not necessarily correspond to human judgments. In this work, we propose RefReward-SR, a low-resolution (LR) reference-aware reward model for preference-aligned SR. Instead of relying on GT supervision or NR evaluation, RefReward-SR assesses high-resolution (HR) reconstructions conditioned on their LR inputs, treating the LR image as a semantic anchor. Leveraging the visual-linguistic priors of a Multimodal Large Language Models (MLLM), it evaluates semantic consistency and plausibility in a reasoning-aware manner. To support this paradigm, we construct RefSR-18K, the first large-scale LR-conditioned preference dataset for SR, providing pairwise rankings based on LR-HR consistency and HR naturalness. We fine-tune the MLLM with Group Relative Policy Optimization (GRPO) using LR-conditioned ranking rewards, and further integrate GRPO into SR model training with RefReward-SR as the core reward signal for preference-aligned generation. Extensive experiments show that our framework achieves substantially better alignment with human judgments, producing reconstructions that preserve semantic consistency while enhancing perceptual plausibility and visual naturalness. Code, models, and datasets will be released upon paper acceptance.

Related papers

• Detail Continuation over a Trustworthy Coarse Scale for Autoregressive Super-Resolution [11.303661953283894]
  Hallucination remains a persistent challenge in generative super-resolution.<n> division between preserving reliable coarse-scale information and restoring more uncertain fine details is often handled implicitly.<n>We propose textbfK2N, which reformulates VAR-based SR from full-path generation into a $k$-to-$N$ detail continuation process.
  arXiv  Detail & Related papers  (2026-08-03T07:29:53Z)
• FoA-SR: Faithful or Aesthetic? Profile-Aware Preference Optimization for Real-World Image Super-Resolution [11.365750371241155]
  We propose FoA-SR, a novel preference optimization approach to real-world image super-resolution based on profiles.<n>To achieve this goal, FoA-SR starts with our supervised FLUX.2-based SR adapter (Flux2SR) trained with LR latent conditioning, flow matching, and image-space reconstruction losses.<n>Experiments on RealSR and DIV2K show that FoA-SR can steer the same SR adapter towards distinct restoration objectives.
  arXiv  Detail & Related papers  (2026-06-09T00:44:59Z)
• DegBins: Degradation-Driven Binning for Depth Super-Resolution [57.61783461543096]
  DegBins is a novel DSR framework that leverages degradation-driven binning to adaptively enhance residual modeling.<n>DegBins consistently outperforms existing state-of-the-art methods in terms of accuracy, robustness, and generalization.
  arXiv  Detail & Related papers  (2026-05-10T16:17:11Z)
• Optimizing RAG Rerankers with LLM Feedback via Reinforcement Learning [47.963236269307735]
  Rerankers play a pivotal role in refining retrieval results for Retrieval-Augmented Generation.<n>Current reranking models are typically optimized on static human annotated relevance labels in isolation, decoupled from the downstream generation process.<n>We introduce ReRanking Preference Optimization (RRPO), a reinforcement learning framework that directly aligns reranking with the LLM's generation quality.
  arXiv  Detail & Related papers  (2026-04-02T14:19:47Z)
• LucidNFT: LR-Anchored Multi-Reward Preference Optimization for Generative Real-World Super-Resolution [21.290660354883595]
  Preference-based reinforcement learning (RL) is a natural fit because each LR input yields a rollout group of candidates to compare.<n>We propose LucidNFT, a multi-reward RL framework for flow-matching Real-ISR.<n>Experiments show that LucidNFT consistently improves strong flow-based Real-ISR baselines.
  arXiv  Detail & Related papers  (2026-03-06T06:30:34Z)
• RASR: Retrieval-Augmented Super Resolution for Practical Reference-based Image Restoration [41.50858324338511]
  RefSR improves upon Single Image Super Resolution (SISR) by leveraging high-quality reference images to enhance texture fidelity and visual realism.<n>Existing RefSR approaches rely on manually curated target-reference image pairs.<n>We introduce Retrieval-Augmented Super Resolution (RASR), a new and practical RefSR paradigm that automatically retrieves semantically relevant high-resolution images from a reference database.
  arXiv  Detail & Related papers  (2025-08-13T03:05:20Z)
• REBEL: Reinforcement Learning via Regressing Relative Rewards [59.68420022466047]
  We propose REBEL, a minimalist RL algorithm for the era of generative models.<n>In theory, we prove that fundamental RL algorithms like Natural Policy Gradient can be seen as variants of REBEL.<n>We find that REBEL provides a unified approach to language modeling and image generation with stronger or similar performance as PPO and DPO.
  arXiv  Detail & Related papers  (2024-04-25T17:20:45Z)
• Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning [45.13580581290495]
  This work introduces a novel "Low-Res Leads the Way" (LWay) training framework to enhance the adaptability of SR models to real-world images. Our approach utilizes a low-resolution (LR) reconstruction network to extract degradation embeddings from LR images, merging them with super-resolved outputs for LR reconstruction. Our training regime is universally compatible, requiring no network architecture modifications, making it a practical solution for real-world SR applications.
  arXiv  Detail & Related papers  (2024-03-05T02:29:18Z)
• ICF-SRSR: Invertible scale-Conditional Function for Self-Supervised Real-world Single Image Super-Resolution [60.90817228730133]
  Single image super-resolution (SISR) is a challenging problem that aims to up-sample a given low-resolution (LR) image to a high-resolution (HR) counterpart. Recent approaches are trained on simulated LR images degraded by simplified down-sampling operators. We propose a novel Invertible scale-Conditional Function (ICF) which can scale an input image and then restore the original input with different scale conditions.
  arXiv  Detail & Related papers  (2023-07-24T12:42:45Z)
• RRSR:Reciprocal Reference-based Image Super-Resolution with Progressive Feature Alignment and Selection [66.08293086254851]
  We propose a reciprocal learning framework to reinforce the learning of a RefSR network. The newly proposed module aligns reference-input images at multi-scale feature spaces and performs reference-aware feature selection. We empirically show that multiple recent state-of-the-art RefSR models can be consistently improved with our reciprocal learning paradigm.
  arXiv  Detail & Related papers  (2022-11-08T12:39:35Z)
• MASA-SR: Matching Acceleration and Spatial Adaptation for Reference-Based Image Super-Resolution [74.24676600271253]
  We propose the MASA network for RefSR, where two novel modules are designed to address these problems. The proposed Match & Extraction Module significantly reduces the computational cost by a coarse-to-fine correspondence matching scheme. The Spatial Adaptation Module learns the difference of distribution between the LR and Ref images, and remaps the distribution of Ref features to that of LR features in a spatially adaptive way.
  arXiv  Detail & Related papers  (2021-06-04T07:15:32Z)
• Robust Reference-based Super-Resolution via C2-Matching [77.51610726936657]
  Super-Resolution (Ref-SR) has recently emerged as a promising paradigm to enhance a low-resolution (LR) input image by introducing an additional high-resolution (HR) reference image. Existing Ref-SR methods mostly rely on implicit correspondence matching to borrow HR textures from reference images to compensate for the information loss in input images. We propose C2-Matching, which produces explicit robust matching crossing transformation and resolution.
  arXiv  Detail & Related papers  (2021-06-03T16:40:36Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.