Generative Models for Low-Level Vision: Advances, Applications and Symmetry-Driven Design

Dear Colleagues,

This Special Issue aims to gather cutting-edge research on the design, optimization and real-world application of generative models in low-level vision tasks, and to explore the intrinsic connection between generative model design and the symmetry principle in computer vision. Its scope covers the theoretical innovation of generative models (diffusion models, GANs, etc.) for low-level vision, task-specific model optimization, cross-domain application expansion, and the integration of symmetry prior into generative model architecture. The Special Issue will showcase novel methods that address the key challenges of low-level vision (e.g., complex image degradation, detail loss, low efficiency) via generative models and highlight how symmetry—an inherent geometric property of visual data—boosts the performance of generative models in low-level vision tasks. Symmetry is embedded in the core design of generative models for low-level vision: for example, symmetric decoders in latent diffusion models ensure scale-consistent image super-resolution, reflection symmetry priors guide accurate shape completion and image inpainting, and geometric symmetry constraints reduce the ambiguity of generative model sampling in image restoration. By incorporating symmetry into generative model training and inference, the interpretability and generalization ability of models for low-level vision tasks (e.g., super-resolution, deblurring, inpainting) are significantly enhanced, enabling more robust and high-fidelity visual data processing.

We are pleased to invite you to contribute to this Special Issue focusing on Generative Models for Low-Level Vision: Advances, Applications and Symmetry-Driven Design. Low-level vision is a fundamental research direction of computer vision, focusing on restoring and enhancing low-quality visual data degraded by complex real-world factors (e.g., noise, blur, low light, occlusion), and its research results are widely applied in medical imaging, remote sensing, autonomous driving, and consumer electronics. Traditional low-level vision methods rely on handcrafted priors and variational optimization, which struggle to handle complex and diverse image degradation scenarios. In recent years, deep generative models (e.g., diffusion models, generative adversarial networks, variational autoencoders) have emerged as a powerful tool for low-level vision, thanks to their strong capability of learning complex data distributions and generating high-fidelity, diverse visual content. Among them, diffusion models have become the mainstream framework for low-level vision tasks due to their excellent performance in image super-resolution, deblurring, inpainting, and low-light enhancement. However, current generative models for low-level vision still face critical challenges: high computational cost, detail loss in generation, poor generalization to unseen degradation, and lack of interpretability in model design. Meanwhile, symmetry—a ubiquitous geometric property in natural and artificial visual data—has long been a key structural constraint in computer vision. The integration of symmetry priors into generative model design has been proven to effectively solve the above challenges, such as improving the scale consistency of super-resolution models, reducing the sampling ambiguity of inpainting models, and enhancing the geometric rationality of shape completion models. This research direction not only promotes the innovation of generative model theory for low-level vision but also provides a new paradigm for the practical application of low-level vision technology, with important scientific significance and engineering value.

This Special Issue aims to provide a high-quality academic platform for researchers in the fields of computer vision, machine learning, and image processing to share the latest research achievements, technical breakthroughs, and practical experience of generative models in low-level vision. We seek to collect innovative research that addresses the core challenges of generative models in low-level vision, and to explore the deep integration of symmetry principles and generative model design. The Special Issue also aims to bridge the gap between the theoretical research of generative models and practical application of low-level vision, and to promote the development of low-level vision technology towards high-efficiency, high-fidelity, and strong generalization. This subject is highly consistent with the journal’s scope of focusing on computer vision, machine learning, and image processing theory and applications: it covers the fundamental algorithm innovation of generative models, the in-depth research of low-level vision tasks, and the cross-disciplinary application of visual technology, and the exploration of symmetry-driven model design also conforms to the journal’s pursuit of in-depth research on the intrinsic principles of visual computing. The scope of this Special Issue is neither overly broad nor narrow: it focuses on the specific intersection of generative models and low-level vision and further refines the research direction by taking symmetry as a core design perspective, which is conducive to gathering focused and in-depth research results. We aim to collect at least 10 high-quality articles, and look forward to compiling the Special Issue into a book form to disseminate cutting-edge research results in this field to a wider academic and industrial community. In this Special Issue, original research articles and reviews are welcome. Research areas may include (but are not limited to) the following:

• Novel generative model architectures (diffusion models, GANs, VAEs) for classic low-level vision tasks (super-resolution, deblurring, denoising, inpainting, low-light enhancement, dehazing)
• Symmetry-driven design and optimization of generative models for low-level vision (e.g., symmetric decoders, reflection symmetry priors, geometric symmetry constraints)
• Efficient generative models for low-level vision (lightweight design, fast sampling, low-computation optimization)
• Cross-domain application of generative models in low-level vision (medical imaging reconstruction, remote sensing image enhancement, video low-level processing)
• Evaluation metrics and benchmark datasets for generative model-based low-level vision methods
• Fusion of multi-modal information and generative models for low-level vision—solving real-world low-level vision challenges (e.g., complex mixed degradation) via generative models
• Survey and perspective of generative models in low-level vision and symmetry-driven visual computing

We look forward to receiving your contributions.

Prof. Dr. Xixi Jia
Guest Editor

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 250 words) can be sent to the Editorial Office for assessment.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a single-blind peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Symmetry is an international peer-reviewed open access monthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 2400 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.

• generative models
• low-level vision
• diffusion models
• image restoration
• symmetry prior
• super-resolution
• image inpainting
• computer vision