Problem
Research questions and friction points this paper is trying to address.
Underwater Image Enhancement
Resource-Constrained Devices
Real-time Deployment
Model Compression
Physics-Guided Spectral Distillation for Underwater Image Enhancement on Resource-Constrained Devices
🤖 AI Summary
This study addresses the challenge of deploying underwater image enhancement models in real time on resource-constrained devices by proposing a physics-guided spectral distillation method. For the first time, physical degradation modeling is incorporated into a knowledge distillation framework. Specifically, the Haar discrete wavelet transform is employed to separate multi-scale frequency band information, while band-specific objective functions are designed using degradation-aware weights and reliability masks to enable efficient knowledge transfer to lightweight models. Experimental results demonstrate that the proposed approach achieves a favorable balance between enhanced image quality and inference speed across multiple datasets, significantly improving downstream detection performance. Furthermore, successful deployment on an in-house developed remotely operated vehicle (ROV) platform validates its practical applicability in real-world scenarios.
📝 Abstract
Underwater image enhancement is crucial for improving visual perception in marine applications. Existing underwater image enhancement studies mainly focus on enhancement quality and visual fidelity, while rarely considering real-time deployment capability, which is essential for resource-constrained underwater robots. To this end, we introduce a physics-guided spectral distillation (PSD) method, which reduces model capacity for real-time applications while maintaining the high performance of underwater image enhancement models. To decompose the outputs of teacher and student models, PSD adopts a multilevel Haar discrete wavelet transform. It transfers low-frequency color and illumination information as well as high-frequency structural details through band-specific objectives. Moreover, the distillation process of PSD is degradation-aware. We estimate degradation-aware weights through a physical head and combine them with ground-truth-guided reliability masks to selectively retain valuable teacher guidance. Experiments on the UIEB, LSUI, and EUVP datasets validate the effectiveness of the proposed method. Furthermore, we demonstrate the benefits of enhanced images for downstream perception tasks, including object detection. Deployment on a self-developed ROV further demonstrates its practical applicability in real-world underwater scenarios.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Physics-Guided Spectral Distillation
Underwater Image Enhancement
Haar Discrete Wavelet Transform
Degradation-Aware
Knowledge Distillation
🔎 Similar Papers
No similar papers found.
