Image restoration agent predicts results before fixing photos
Look Before You Restore: Learning to Predict Before Acting for Agentic Image Restoration
Computer Vision and Pattern Recognition
Summary
Images can get damaged in different ways, and fixing them often involves trying many tools without knowing if they will help. The authors built a system called PreActIR that guesses how well each tool will work before actually using it, which saves time and prevents making photos worse. It also checks its own fixes with a verifier that decides to keep or undo the change. This method uses fewer tries and avoids bad corrections, leading to better restored images.
What this means in practice
- •For photo editing software developers: Improve automatic photo repair by predicting the effect of correction tools before applying them to reduce wasted processing time and harmful edits.
- •For digital content managers: Enhance large-scale image restoration workflows by minimizing unnecessary tool executions and ensuring higher fidelity outputs through predictive verification.
Authors
Xu Zhang, Jun Wan, Lefei Zhang, Liangpei Zhang
Abstract
Agents for mixed-degradation image restoration typically learn what a tool does to an image only after running it: failed calls waste computation and harmful outputs go unnoticed until too late. We propose PreActIR, an agentic restoration framework that predicts tool consequences before execution. For each candidate tool, a tool-conditioned world model forecasts changes in residual degradation, image quality, and fidelity, together with the risk of harm and the uncertainty of the prediction. The agent executes the top-ranked candidate, and an independent verifier commits or rolls back its output before the agent replans from the observed state. PreActIR also audits whether performance is limited by the selector or by the candidate space, and studies targeted tool adaptation as a controlled way to expand that space. In a controlled evaluation that fixes the candidate pool, verifier, and budget and employs reference-annotated degradation states, PreActIR uses 42.5% fewer real tool calls and 63.4% fewer rejected calls than reactive, execute-then-evaluate trial and error, while improving Y-PSNR by 0.92 dB.