Adaptive Dynamic Dehazing via Instruction-Driven and Task-Feedback Closed-Loop Optimization for Diverse Downstream Task Adaptation

TL;DR

A novel adaptive dynamic dehazing framework that incorporates a closed-loop optimization mechanism that enables feedback-driven refinement based on downstream task performance and user instruction-guided adjustment during inference, allowing the model to satisfy the specific requirements of multiple downstream tasks without retraining.

Abstract

In real-world vision systems,haze removal is required not only to enhance image visibility but also to meet the specific needs of diverse downstream tasks.To address this challenge,we propose a novel adaptive dynamic dehazing framework that incorporates a closed-loop optimization mechanism.It enables feedback-driven refinement based on downstream task performance and user instruction-guided adjustment during inference,allowing the model to satisfy the specific requirements of multiple downstream tasks without retraining.Technically,our framework integrates two complementary and innovative mechanisms: (1)a task feedback loop that dynamically modulates dehazing outputs based on performance across multiple downstream tasks,and (2) a text instruction interface that allows users to specify high-level task preferences.This dual-guidance strategy enables the model to adapt its dehazing behavior after training,tailoring outputs in real time to the evolving needs of multiple tasks.Extensive experiments across various vision tasks demonstrate the strong effectiveness,robustness,and generalizability of our approach.These results establish a new paradigm for interactive,task-adaptive dehazing that actively collaborates with downstream applications.

Figures (11)

• Figure 1: Comparison between existing methods and ours.
• Figure 2: Overview of the proposed method. The method constructs a closed-loop regulation mechanism jointly guided by semantic task instructions and task performance feedback. It leverages the semantic features of text instructions extracted by BERT, the initial dehazed images, and feedback from downstream tasks to collaboratively adjust dehazing features via the IGM and TFGA modules, enabling adaptive optimization across diverse downstream scenario.
• Figure 3: Structure of the TFGA.
• Figure 4: Structure of the IGM.
• Figure 5: Visual comparison with state-of-the-art methods on Setting 1. Each part includes two rows: dehazing results (rows 1) and corresponding downstream task outputs (rows 2). Input images are taken from ADE20K, COCO, and KITTI.