OARS: Process-Aware Online Alignment for Generative Real-World Image Super-Resolution

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhao, Shijie, Zhang, Xuanyu, Chen, Bin, Li, Weiqi, Xing, Qunliang, Zhang, Kexin, Wang, Yan, Li, Junlin, Zhang, Li, Zhang, Jian, Xue, Tianfan
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910051712630784
author Zhao, Shijie
Zhang, Xuanyu
Chen, Bin
Li, Weiqi
Xing, Qunliang
Zhang, Kexin
Wang, Yan
Li, Junlin
Zhang, Li
Zhang, Jian
Xue, Tianfan
author_facet Zhao, Shijie
Zhang, Xuanyu
Chen, Bin
Li, Weiqi
Xing, Qunliang
Zhang, Kexin
Wang, Yan
Li, Junlin
Zhang, Li
Zhang, Jian
Xue, Tianfan
contents Aligning generative real-world image super-resolution models with human visual preference is challenging due to the perception--fidelity trade-off and diverse, unknown degradations. Prior approaches rely on offline preference optimization and static metric aggregation, which are often non-interpretable and prone to pseudo-diversity under strong conditioning. We propose OARS, a process-aware online alignment framework built on COMPASS, a MLLM-based reward that evaluates the LR to SR transition by jointly modeling fidelity preservation and perceptual gain with an input-quality-adaptive trade-off. To train COMPASS, we curate COMPASS-20K spanning synthetic and real degradations, and introduce a three-stage perceptual annotation pipeline that yields calibrated, fine-grained training labels. Guided by COMPASS, OARS performs progressive online alignment from cold-start flow matching to full-reference and finally reference-free RL via shallow LoRA optimization for on-policy exploration. Extensive experiments and user studies demonstrate consistent perceptual improvements while maintaining fidelity, achieving state-of-the-art performance on Real-ISR benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2603_12811
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle OARS: Process-Aware Online Alignment for Generative Real-World Image Super-Resolution
Zhao, Shijie
Zhang, Xuanyu
Chen, Bin
Li, Weiqi
Xing, Qunliang
Zhang, Kexin
Wang, Yan
Li, Junlin
Zhang, Li
Zhang, Jian
Xue, Tianfan
Computer Vision and Pattern Recognition
Aligning generative real-world image super-resolution models with human visual preference is challenging due to the perception--fidelity trade-off and diverse, unknown degradations. Prior approaches rely on offline preference optimization and static metric aggregation, which are often non-interpretable and prone to pseudo-diversity under strong conditioning. We propose OARS, a process-aware online alignment framework built on COMPASS, a MLLM-based reward that evaluates the LR to SR transition by jointly modeling fidelity preservation and perceptual gain with an input-quality-adaptive trade-off. To train COMPASS, we curate COMPASS-20K spanning synthetic and real degradations, and introduce a three-stage perceptual annotation pipeline that yields calibrated, fine-grained training labels. Guided by COMPASS, OARS performs progressive online alignment from cold-start flow matching to full-reference and finally reference-free RL via shallow LoRA optimization for on-policy exploration. Extensive experiments and user studies demonstrate consistent perceptual improvements while maintaining fidelity, achieving state-of-the-art performance on Real-ISR benchmarks.
title OARS: Process-Aware Online Alignment for Generative Real-World Image Super-Resolution
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.12811