Abstract
Object detection methods under known single degradations have beenextensively investigated. However, existing approaches require prior knowledgeof the degradation type and train a separate model for each, limiting theirpractical applications in unpredictable environments. To address thischallenge, we propose a chain-of-thought (CoT) prompted adaptive enhancer,CPA-Enhancer, for object detection under unknown degradations. Specifically,CPA-Enhancer progressively adapts its enhancement strategy under thestep-by-step guidance of CoT prompts, that encode degradation-relatedinformation. To the best of our knowledge, it's the first work that exploitsCoT prompting for object detection tasks. Overall, CPA-Enhancer is aplug-and-play enhancement model that can be integrated into any genericdetectors to achieve substantial gains on degraded images, without knowing thedegradation type priorly. Experimental results demonstrate that CPA-Enhancernot only sets the new state of the art for object detection but also boosts theperformance of other downstream vision tasks under unknown degradations.