Technology Category
Computer Vision: Multi-modal VisionMachine Learning: Multimodal LearningIntelligent Robots: Multimodal Perception & Sensor Fusion
A Survey on All-in-One Image Restoration: Taxonomy, Evaluation and Future Trends
🤖 AI Summary
To address the poor generalization of conventional single-degradation image restoration methods under realistic scenarios where multiple degradations (e.g., noise, blur, weather artifacts) co-occur, this paper proposes a unified All-in-One Image Restoration (AiOIR) paradigm. We first establish a systematic taxonomy for AiOIR; then introduce a three-dimensional evaluation framework covering prior modeling, generalization capability, and learning paradigms; and finally design an adaptive architecture integrating multi-task learning, meta-learning, degradation-aware attention, and a shared-specialized dual-path network. Extensive benchmarking is conducted on mainstream datasets (Rain13k, RealBlur, DPSR) using PSNR, SSIM, and LPIPS metrics, with open-source method comparisons. Contributions include: (i) the first structured AiOIR survey, (ii) an objective performance benchmark, (iii) an open-source codebase (GitHub), and (iv) identified future directions—scalable architectures, disentangled representations, and dynamic inference.
Technology Category
Application Category
📝 Abstract
Image restoration (IR) refers to the process of improving visual quality of images while removing degradation, such as noise, blur, weather effects, and so on. Traditional IR methods typically target specific types of degradation, which limits their effectiveness in real-world scenarios with complex distortions. In response to this challenge, the all-in-one image restoration (AiOIR) paradigm has emerged, offering a unified framework that adeptly addresses multiple degradation types. These innovative models enhance both convenience and versatility by adaptively learning degradation-specific features while simultaneously leveraging shared knowledge across diverse corruptions. In this review, we delve into the AiOIR methodologies, emphasizing their architecture innovations and learning paradigm and offering a systematic review of prevalent approaches. We systematically categorize prevalent approaches and critically assess the challenges these models encounter, proposing future research directions to advance this dynamic field. Our paper begins with an introduction to the foundational concepts of AiOIR models, followed by a categorization of cutting-edge designs based on factors such as prior knowledge and generalization capability. Next, we highlight key advancements in AiOIR, aiming to inspire further inquiry and innovation within the community. To facilitate a robust evaluation of existing methods, we collate and summarize commonly used datasets, implementation details, and evaluation metrics. Additionally, we present an objective comparison of open-sourced methods, providing valuable insights for researchers and practitioners alike. This paper stands as the first comprehensive and insightful review of AiOIR. A related repository is available at https://github.com/Harbinzzy/All-in-One-Image-Restoration-Survey.
Problem
Research questions and friction points this paper is trying to address.
Surveying all-in-one image restoration methods for multiple degradations
Providing taxonomy and evaluation of AiOIR architectures and strategies
Proposing future research directions for unified image restoration
Innovation
Methods, ideas, or system contributions that make the work stand out.
Unified framework for multiple degradations
Adaptive learning of degradation features
Shared knowledge across diverse corruptions
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
Harbin Institute of Technology
