Abstract:Blind all-in-one image restoration models aim to recover a high-quality image from an input degraded with unknown distortions. However, these models require all the possible degradation types to be defined during the training stage while showing limited generalization to unseen degradations, which limits their practical application in complex cases. In this paper, we propose a simple but effective adaptive blind all-in-one restoration (ABAIR) model, which can address multiple degradations, generalizes well to unseen degradations, and efficiently incorporate new degradations by training a small fraction of parameters. First, we train our baseline model on a large dataset of natural images with multiple synthetic degradations, augmented with a segmentation head to estimate per-pixel degradation types, resulting in a powerful backbone able to generalize to a wide range of degradations. Second, we adapt our baseline model to varying image restoration tasks using independent low-rank adapters. Third, we learn to adaptively combine adapters to versatile images via a flexible and lightweight degradation estimator. Our model is both powerful in handling specific distortions and flexible in adapting to complex tasks, it not only outperforms the state-of-the-art by a large margin on five- and three-task IR setups, but also shows improved generalization to unseen degradations and also composite distortions.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve several key challenges in image restoration (IR), especially the limitations in dealing with unknown and compound degradations. Specifically, the paper attempts to solve the following problems: 1. **Blind restoration problem**: - Existing all - in - one image restoration models need to define all possible degradation types during the training phase, and show limited generalization ability to unseen degradation types in practical applications. This restricts their practical applications in complex situations. - The paper proposes an Adaptive Blind All - in - One Restoration (ABAIR) model, which can handle multiple degradations without knowing the degradation types and can generalize well to unseen degradation types. 2. **Compound degradation processing**: - Actual images often contain multiple compound degradations (for example, moving vehicles in low - light scenes, compressed images taken in foggy days). Existing models usually can only handle a single type of degradation, limiting their effectiveness in practical applications. - The ABAIR model can effectively handle compound degradations by introducing a lightweight degradation estimator and an adapter module, thus improving the applicability of the model in complex scenes. 3. **Extension of new degradation types**: - Current all - in - one methods require all degradation types to be defined during training, making it difficult to add new degradation types without retraining the entire model, which is both time - consuming and computationally expensive. - The ABAIR model achieves flexible expansion by training only a small number of parameters to add new degradation types and retraining the lightweight estimator, without forgetting previously learned tasks. ### Summary The ABAIR model proposed in the paper aims to solve the above problems in the following ways: - **Large - scale pre - training**: Use synthetic degradation data for pre - training to obtain strong initial weights. - **Independent adapters**: Train independent adapter modules for each degradation type to bridge the gap between synthetic data and real - world data. - **Lightweight degradation estimator**: Learn a lightweight degradation estimator to select the most appropriate adapter combination according to the input image, thus achieving blind restoration. Through these methods, the ABAIR model not only achieves results significantly better than existing methods in multiple standard benchmark tests, but also performs well in unseen degradation types and compound degradation scenarios.

Adaptive Blind All-in-One Image Restoration

Efficient Degradation-aware Any Image Restoration

Rdm-Ir: Task-Adaptive Deep Unfolding Network for All-in-One Image Restoration

DRM-IR: Task-Adaptive Deep Unfolding Network for All-In-One Image Restoration

AdaIR: Exploiting Underlying Similarities of Image Restoration Tasks with Adapters

Multi-modal Degradation Feature Learning for Unified Image Restoration Based on Contrastive Learning

AdaIR: Adaptive All-in-One Image Restoration via Frequency Mining and Modulation

Restorer: Removing Multi-Degradation with All-Axis Attention and Prompt Guidance

All-in-one Multi-degradation Image Restoration Network via Hierarchical Degradation Representation

Unified-Width Adaptive Dynamic Network for All-In-One Image Restoration

Perceive-IR: Learning to Perceive Degradation Better for All-in-One Image Restoration

Blind Image Restoration via Fast Diffusion Inversion

Mixed Degradation Image Restoration via Local Dynamic Optimization and Conditional Embedding

Restore Anything with Masks: Leveraging Mask Image Modeling for Blind All-in-One Image Restoration

Toward Efficient Deep Blind RAW Image Restoration

InstantIR: Blind Image Restoration with Instant Generative Reference

PromptIR: Prompting for All-in-One Blind Image Restoration

DiffBIR: Towards Blind Image Restoration with Generative Diffusion Prior

LoRA-IR: Taming Low-Rank Experts for Efficient All-in-One Image Restoration

LIR: A Lightweight Baseline for Image Restoration

Spatially-Adaptive Image Restoration using Distortion-Guided Networks