Self-Evolving Agentic Image Restoration via Deliberate Planning and Intuitive Execution

2026-06-27

AI Summary

This work addresses the challenge of real-world image restoration under complex, coupled degradations, where existing agent-based methods struggle to balance exploration and exploitation due to greedy search strategies and suffer from insufficient information utilization and catastrophic forgetting. To overcome these limitations, the paper formulates restoration as a sequential decision-making problem and proposes a self-evolving agent framework grounded in dual-process theory, integrating an intuitive executor with a deliberative planner. The approach introduces a pruning-aware Monte Carlo tree search for long-horizon reasoning and devises a degradation-aware state fingerprint to drive episodic memory, effectively mitigating forgetting and reducing cold-start costs. Evaluated with a no-reference hybrid reward and multimodal large language model–based assessment, the method achieves state-of-the-art perceptual quality and quantitative performance on both synthetic and real-world benchmarks.

Abstract

Real-world image restoration (IR) remains challenging due to complex and coupled degradations. While recent agentic IR frameworks leverage Large Language Models for flexible tool planning, they face two critical limitations. First, from a search scheme perspective, excessive reliance on greedy strategies fails to balance exploration and exploitation. Second, existing agentic systems underutilize information, exhibiting episodic amnesia. To address these challenges, we propose \textbf{Self-Evolving Agentic Image Restoration (SEAR)}, which formulates restoration as a sequential decision-making problem. Inspired by the dual-process theory, SEAR comprises an Intuitive Executor and a Deliberate Planner, respectively following the fast-thinking \textit{System 1} and slow-thinking \textit{System 2} principles. The Deliberate Planner employs Pruning-Aware Monte Carlo Tree Search for long-horizon reasoning, utilizing a hybrid no-reference reward and a Multimodal Large Language Model (MLLM)-based tournament to prevent metric exploitation. Complementarily, the Intuitive Executor leverages a self-evolving episodic memory indexed by degradation-aware state fingerprints. This mechanism distills expensive search trajectories into adaptive expertise, overcoming episodic amnesia while progressively amortizing cold-start exploration costs through memory reuse. Extensive experiments on synthetic and real-world benchmarks demonstrate its strong perceptual and quantitative performance.

Problem

• image restoration
• agentic systems
• exploration-exploitation tradeoff
• episodic amnesia
• complex degradations

Innovation

• Self-Evolving Agentic System
• Deliberate Planning
• Intuitive Execution
• Monte Carlo Tree Search
• Episodic Memory