Perception Models in Harsh Domains: Detection and Depth in Underwater Images

Gabriel De Lima Anastacio, Emanuel C. Silva, Eduardo Lawson da Silva, André Luiz Ditzel Burger, Stephanie Loi Brião, Tatiana Taís Schein, Paulo L. J. Drews-Jr

Keywords: Harsh scenarios, Underwater images, Robotics perception, Object Detection, Depth-Estimation, Computer vision tasks

TL;DR: Evaluation of depth estimation and object detection models in underwater environments, highlighting trade-offs between robustness and accuracy.

Abstract: Depth estimation and object detection are relevant tasks in computer vision. This work presents a comparative analysis of the perception models for the mentioned tasks in harsh underwater scenarios. In underwater applications such as autonomous navigation, environmental monitoring, and infrastructure inspection, the image degradation can result from light attenuation, scattering, and turbidity. Addressing these challenges requires robust perception models that operate in the constrained conditions, motivating in the evaluation of state-of-the-art approaches. Depth estimation is evaluated using the Marigold, Depth Anything V2, and Depth Anything V3 models. Furthermore, for object detection, architectures such as YOLOv8, YOLOv9, YOLOv10, YOLOv11, YOLOv26, and RF-DETR are utilized, as well as specialized approaches, FeatEnHancer, AMSP-UOD, AquaFeat, and AquaFeat+. The quantitative and qualitative analysis of models' performance and of insights, integrated geometric and semantic information for the perception of robotic systems in ocean exploration.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Paper Acceptance: No

Submission Number: 34