Simulated Forward-Looking Sonar Datasets of Mountainous Terrain and an Underwater UXO Target

The Sonar2DGS Dataset is a multi-scene dataset designed for research on forward-looking imaging sonar scene representation, 3D reconstruction, and novel-view synthesis. It comprises a mountain scene, a UXO scene, two real-world unexploded ordnance acquisition sequences, and four robustness experiment subsets, providing 586 8-bit grayscale sonar images with sonar-to-world pose annotations. The dataset also includes imaging ranges, horizontal and vertical fields of view, normalization parameters, predefined data splits, and optional point clouds. Its robustness subsets cover additive noise, multiplicative noise, pose perturbations, and range-based normalization, enabling systematic evaluation of reconstruction quality and generalization performance for sonar Gaussian Splatting, radiance fields, and related 3D vision methods under varying imaging conditions and sources of error.