Why Domain Matters: A Preliminary Study of Domain Effects in Underwater Object Detection

TL;DR

This paper introduces a new framework for defining underwater image domains based on measurable physical factors, improving the analysis of domain shift effects on object detection models.

Contribution

It proposes a physically meaningful domain labeling framework that captures intrinsic scene factors, enabling better domain-specific evaluation and failure analysis in underwater detection.

Findings

Systematic variations across domain factors were observed.

The framework revealed hidden failure modes in detection models.

Validation on public datasets demonstrated the framework's effectiveness.

Abstract

Domain shift, where deviations between training and deployment data distributions degrade model performance, is a key challenge in underwater environments. Existing benchmarks testing performance for underwater domain shift simulate variability through synthetic style transfer. This fails to capture intrinsic scene factors such as visibility, illumination, scene composition, or acquisition factors, limiting analysis of real-world effects. We propose a labeling framework that defines underwater domains using measurable image, scene, and acquisition characteristics. Unlike prior benchmarks, it captures physically meaningful factors, enabling semantically consistent image grouping and supporting domain-specific evaluation of detection performance including failure analysis. We validate this on public datasets, showing systematic variations across domain factors and revealing hidden failure modes.