Abstract / 摘要

• Generative image restoration and enhancement is currently at a critical turning point, shifting from “pixel completion” toward “semantic generation.” The integration of diffusion models, flow-based models, and multimodal large models has enabled unprecedented detail reconstruction capabilities in extreme degradation scenarios. However, it has also triggered a redefinition of what “realness” means in visual data. When models actively hallucinate missing information, the boundary between hallucination and restoration becomes increasingly blurred, and the trade-off between fidelity and perceptual quality demands systematic re-examination.
• This forum aims to build an interdisciplinary platform for academic exchange, bringing together algorithm researchers, optical engineers, and related experts. Discussions will focus on four core themes: controllability of generative restoration, physical consistency constraints, lightweight deployment, and subjective evaluation frameworks. The forum will particularly emphasize robust solutions for real-world applications such as low-light enhancement, underwater image enhancement, remote sensing image restoration, and historical image recovery. It also advocates for the development of open and reproducible benchmarking standards for generative restoration, providing both theoretical foundations and practical pathways for trustworthy next-generation intelligent imaging systems.

生成式图像复原与增强技术正处于从“像素填充”向“语义生成”跃迁的关键转折期，扩散模型、流模型与多模态大模型的融合，使算法在极端退化场景下展现出前所未有的细节再生能力，但也同步引发了关于“真实”定义的重构。当模型主动补全缺失信息时，幻觉与复原的边界变得模糊，保真度与感知质量的矛盾亟待系统性审视。本分论坛旨在搭建跨学科对话平台，汇聚算法研究者、光学工程师等相关专家学者，围绕生成式复原的可控性、物理一致性约束、轻量化部署及主观评价体系四大核心议题展开深度研讨，重点推动面向低光照增强、水下图像增强、遥感影像复原、历史影像修复等场景的鲁棒解决方案，同时倡导建立开放、可复现的生成式复原评测基准，为下一代智能影像系统的可信增强提供理论锚点与实践路径。

Topics 主题征稿范围

The following topics are within the scope of this special session, but are not limited to: 以下是该分论坛征稿范围，但不限于：

• Video and Image Denoising, Deblurring, and Super-Resolution Reconstruction based on Diffusion Models, GANs, and Flow Models
  基于扩散模型、GAN、流模型等的视频图像去噪、去模糊、超分辨率重建；
• Image Restoration under Extreme Degradation Conditions such as Low-Light, High-Noise, and Adverse Weather
  低光照、强噪声、恶劣天气等极端退化场景下的图像复原技术；
• Video/Image Restoration, Inpainting, Completion, and Visual Quality Enhancement
  视频图像修复、补全、以及视觉质量增强；
• Hybrid Restoration Frameworks Integrating Imaging Physics Models and Generative Priors
  融合成像物理模型与生成先验的混合复原框架；
• Controllable Generative Image Restoration
  可控生成式图像复原；
• Joint Optimization of Visual Enhancement and Resolution in Scene Restoration
  场景恢复中的视觉增强与分辨率联合优化；
• Multimodal and Cross-Sensor Image Fusion and Synthesis
  多模态/跨传感器图像融合与合成；
• Generative Reconstruction, Artifact Correction, and Data Augmentation in Medical Imaging
  医学影像中的生成式重建、伪影校正与数据增广；
• Remote Sensing Image Enhancement, Cloud/Haze Removal, and Land Cover Reconstruction
  遥感影像增强、云/霾去除与地物重建；
• Compression, Distillation, and Real-Time Deployment of Generative Restoration Models
  生成式复原模型的压缩、蒸馏与实时化部署；
• Lightweight Generative Model Design and Inference Acceleration for Mobile and Edge Devices
  面向移动端/边缘设备的轻量化生成模型设计与推理加速；
• Fidelity–Perceptual Quality Trade-off and Subjective/Objective Evaluation of Generative Restoration
  生成式复原的保真度-感知质量平衡与主观/客观评价；
• Provenance, Forgery Detection, and Security Protection Techniques for AIGC Visual Content
  AIGC视觉内容的溯源、鉴伪与安全防护技术；
• Digital Restoration of Cultural Heritage and Enhancement of Historical Imagery
  文化遗产数字化修复与历史影像增强；
• Generative Image Restoration for Underwater, Industrial Defects, and Autonomous Driving Scenarios
  水下/工业缺陷/自动驾驶场景的生成式影像复原。