AdaPrompt-IR: Adaptive learning to perceive degradation semantic and prompting for all-in-one image restoration

Image restoration is a critical task aimed at recovering high-quality, clean images from their degraded counterparts. While deep learning-based methods have made significant strides in this area, they still struggle with generalizing across a wide range of degradation types and levels. This limitation restricts their practical applicability, as it often requires training separate models for each degradation type and knowing the specific degradation in advance to apply the appropriate model. In this work, we introduce AdaPrompt-IR, an adaptive all-in-one image restoration network that combines degradation semantic mining and prompt learning. Our approach leverages a degradation representation codebook to effectively decompose and capture various degradation types. Building upon this foundation, we propose a degradation activation module, integrated with a dual-branch architecture, which explicitly facilitates the restoration of missing degradation semantics. Additionally, we incorporate degradation-specific information as prompts to implicitly guide the restoration process. Extensive experiments demonstrate that our method achieves state-of-the-art performance across multiple image restoration tasks, including image denoising, dehazing, deraining, motion deblurring, and low-light image enhancement. Our code is available at https://github.com/hagerwang/AdaPromote-IR.