Object pose estimation underwater allows an autonomous system to perform tracking and intervention tasks. Nonetheless, underwater target pose estimation is remarkably challenging due to, among many factors, limited visibility, light scattering, cluttered environments, and constantly varying water conditions. An approach is to employ sonar or laser sensing to acquire 3D data, however, the data is not clear and the sensors expensive. For this reason, the community has focused on extracting pose estimates from RGB input. In this work, we propose an approach that leverages 2D object detection to reliably compute 6D pose estimates in different underwater scenarios. We test our proposal with 4 objects with symmetrical shapes and poor texture spanning across $33{,}920$ synthetic and 10 real scenes. All objects and scenes are made available in an open-source dataset that includes annotations for object detection and pose estimation. When benchmarking against similar end-to-end methodologies for 6D object pose estimation, our pipeline provides estimates that are $\sim \!\!\!8$ % more accurate. We also demonstrate the real-world usability of our pose estimation pipeline on an underwater robotic manipulator in a reaching task.

In recent years, data-driven approaches have become increasingly important in industrial computer vision applications, particularly for 6-Degrees-of-Freedom (6-DoF) object pose estimation. However, benchmark datasets may unintentionally introduce biases that affect the reliability of learned models. In this work, we investigate the shortcut bias induced by fiducial ArUco markers in the widely used Linemod dataset. Although such markers are typically absent in real industrial environments, they introduce unintended visual cues that neural networks tend to exploit. As a result, model selection based on state-of-the-art benchmarks can be biased, since the reported performance often reflects reliance on these shortcuts rather than on robust feature extraction. Using saliency map analysis, we show that often a large portion of the model’s attention is concentrated on these markers, revealing the presence of a shortcut that artificially boosts pose estimation performance. To mitigate this issue, we propose a data augmentation pipeline based on generative AI techniques that removes the markers and replaces the background with more realistic synthesized scenes. Experimental results indicate a noticeable drop in performance when models trained on the original Linemod dataset are evaluated in ArUco-free environments, confirming the presence of background-induced biases. Training with the proposed generative-swapped dataset leads to improved robustness and better generalization to unseen scenarios, although it does not fully eliminate the problem. Overall, the results highlight the impact of background-related biases in pose estimation benchmarks and suggest that the proposed augmentation strategy represents a practical and scalable step toward developing more reliable 6-DoF pose estimation systems for industrial applications, while leaving room for further improvements.

Accurate mapping of agricultural field boundaries is essential for the efficient operation of agriculture. Automatic extraction from high-resolution satellite imagery, supported by computer vision techniques, can avoid costly ground surveys. In this paper, we present a pipeline for field delineation based on the Segment Anything Model (SAM), introducing a fine-tuning strategy to adapt SAM to this task. In addition to using published datasets, we describe a method for acquiring a complementary regional dataset that covers areas beyond current sources. Extensive experiments assess segmentation accuracy and evaluate the generalization capabilities. Our approach provides a robust baseline for automated field delineation. The new regional dataset, known as ERAS, is now publicly available.