Underwater 3D mapping improved with scalable Gaussian splatting technique
OceanXL: Large-scale Underwater 3D Gaussian Splatting via Block Partitioning and Adaptive Pruning
Computer Vision and Pattern Recognition
Summary
Creating 3D pictures of underwater scenes is hard because water distorts light and it’s tricky to cover large areas. The paper presents OceanXL, a method that breaks big underwater areas into smaller blocks to build detailed 3D images more efficiently. It also smartly removes unnecessary details to keep the 3D models small without losing quality. The authors tested OceanXL on several large underwater environments and showed it works better than previous methods in handling large scenes while keeping good visual results.
What this means in practice
- •For marine survey teams: Create detailed 3D maps of large underwater areas faster and with less computing resources than before.
- •For virtual reality developers: Produce high-quality underwater scenes that run in real time for immersive VR experiences with efficient model sizes.
Authors
Haoran Wang, Shaoyu Cai, Adrian Azzarelli, Zhuodong Jiang, Guoxi Huang, Eng Tat Khoo, Brett Seymour, Fan Zhang, David Bull, Nantheera Anantrasirichai
Abstract
Underwater 3D reconstruction is critical for marine exploration, ecological monitoring, and subsea infrastructure inspection, yet remains challenging at large scale due to light attenuation, scattering, and limited capture coverage. While 3D Gaussian Splatting (3DGS) enables high-quality real-time rendering, its application to large underwater scenes is constrained by high memory consumption and inefficient optimization over extensive areas. We propose OceanXL, a fast and scalable 3DGS-based framework for large-scale underwater reconstruction. OceanXL adopts a divide-and-conquer strategy, partitioning scenes into spatially coherent blocks to enable efficient optimization while preserving global geometric consistency. We further introduce an adaptive pruning scheme tailored to underwater conditions that removes redundant primitives, producing compact representations without sacrificing visual fidelity. Together, these components improve training efficiency and rendering performance for large scenes. We also introduce a large-scale underwater dataset covering diverse marine environments. Experiments on five large-scale scenes demonstrate favorable scalability, compactness, and efficiency--quality trade-offs over large-scene baselines. Controlled comparisons on the small-scale SeaThru-NeRF dataset further show competitive reconstruction quality with substantially smaller model sizes than underwater-specific methods.