QuReC: All-in-One Image Restoration with Query-Specific Guidance and Local-Global Response Calibration

2026-07-16 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

QuReC is a unified framework designed for all-in-one image restoration, addressing the limitations of existing methods that struggle with spatially heterogeneous or mixed degradations within a single image. Published on 2026-07-16, this framework overcomes the inadequacy of image-level prompts or shared guidance by providing fine-grained, spatially adaptive restoration. QuReC integrates a Degradation-Guided Query Reconstruction Module (DQRM), which matches spatial queries against a degradation prototype space to create query-specific degradation-aware representations, enhanced by a weakly supervised prototype matching learning strategy for stability. Additionally, it includes a Local-Global Response Calibration Module (LGRCM) that performs dual-branch aggregation and calibrates responses with learnable priors, ensuring reliable feature aggregation and coordinated local detail and global context modeling. Experiments demonstrate QuReC's superior performance across multiple all-in-one image restoration benchmarks, with its code available on GitHub.

Key takeaway

For computer vision engineers developing robust image restoration models, QuReC offers a significant advancement. You should consider integrating its spatially adaptive guidance and local-global response calibration to handle complex, mixed image degradations more effectively. This approach improves restoration accuracy beyond traditional methods, especially for heterogeneous corruptions. Explore the released code to implement this unified framework and enhance your model's performance across diverse degradation types.

Key insights

QuReC provides spatially adaptive, fine-grained image restoration by integrating query-specific degradation guidance with local-global context calibration.

Principles

• Spatially adaptive guidance is vital for heterogeneous image degradations.
• Reliable local-global context aggregation enhances restoration accuracy.
• Weakly supervised prototype matching stabilizes degradation-aware representation.

Method

QuReC employs a DQRM for query-specific degradation-aware representation via prototype matching and an LGRCM for local-global dual-branch aggregation with learnable prior calibration.

In practice

• Apply QuReC for unified handling of diverse image corruptions.
• Utilize the released code for all-in-one image restoration tasks.

Topics

• All-in-One Image Restoration
• Spatially Adaptive Guidance
• Degradation Prototypes
• Local-Global Context Modeling
• Computer Vision
• Deep Learning Frameworks

Code references

• zhoushen1/QuReC

Best for: Research Scientist, AI Scientist, Computer Vision Engineer, Machine Learning Engineer

Related on AIssential

• The Second LoViF 2026 Challenge on Real-World All-in-One Image Restoration: Methods and Results — The LoViF 2026 Challenge established a benchmark for unified image restoration across diverse real-world degradations.
• CLUIE: Clustering-Aware Recurrent Propagation with Local Structural Compensation for Underwater Image Enhancement — Content-adaptive recurrent propagation and local structural compensation enhance underwater images by addressing heterogeneous degradation.
• RPC-GS: Gaussian Splatting with native RPC Rendering for Satellite Imagery — Native RPC integration in Gaussian Splatting significantly improves satellite imagery 3D reconstruction accuracy.
• Bridging the Training-Deployment Gap: Gated Encoding and Multi-Scale Refinement for Efficient Quantization-Aware Image Enhancement — QAT and hierarchical architecture improve mobile image enhancement quality and efficiency.
• Any Resolution Any Geometry: From Multi-View To Multi-Patch — URGT refines high-resolution depth and normal maps using a multi-patch transformer with global coherence.
• When Recovery Matters: The Blind Spot of Surrogate Privacy in MLLM Editing — Surrogate-based MLLM image editing requires explicit recovery mechanisms to transfer edits back to private source images.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.