IEUOD: Improving Underwater Object Detection via Shallow Feature Guidance from Underwater Image Enhancement Models

Underwater object detection (UOD) is crucial for autonomous underwater robots in exploration and monitoring. However, the detection accuracy of advanced object detection models declines in severely degraded underwater scenes. While underwater image enhancement (UIE) is often used to improve visual quality prior to detection, we find that human and machine visual perception are different. Most enhanced results impair detection accuracy. In this paper, we propose Image Enhancement Guided Underwater Object Detection (IEUOD), a novel training framework that uses enhanced images as supervisory signals rather than direct inputs. A Shallow Feature Guidance (SFG) module aligns shallow features of the detector with those from enhanced images, transferring perceptual priors to improve feature restoration without altering semantic content. The UIE branch and SFG module are discarded during inference, incurring zero computational overhead. Experiments show that using different UIE methods within the IEUOD framework can effectively improve the detection accuracy of multiple detectors. IEUOD achieves a superior accuracy-efficiency trade-off, making it ideal for deployment on resource-constrained underwater platforms. Code and models about IEUOD are available at: https://github.com/qkkk22/IEUOD.