Search papers, labs, and topics across Lattice.
This study addresses the challenge of domain shift in underwater object detection by introducing a novel labeling framework that categorizes underwater domains based on measurable image, scene, and acquisition characteristics. By moving beyond synthetic style transfer, the framework enables a more nuanced understanding of intrinsic factors like visibility and illumination, which are critical for real-world applications. Validation on public datasets reveals systematic variations in detection performance across these domain factors, uncovering previously hidden failure modes that could inform future model improvements.
Systematic variations in underwater object detection performance reveal hidden failure modes tied to intrinsic scene factors, challenging existing benchmarks based on synthetic style transfer.
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.
