GOFish: Full Field-of-View Gaussian Splatting and Mesh Extraction
Authors: Allen Tu, Meher Gitika Karumuri, Shifan Zhu, Hailing Wang, Yen-Liang Lin, David Held, Matthias Zwicker, Tom Goldstein
Subjects: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR); Robotics (cs.RO)
Cite as: arXiv:2610.37818 [cs.CV] — Submitted 27 Oct 2026
Project Page: https://3dgofish.github.io/
Overview
Robots, autonomous vehicles, and consumer 360° cameras rely on fisheye lenses to capture wide fields of view (FoVs). Building simulation-ready digital twins for these platforms requires two things: photorealistic rendering across the full FoV and an aligned collision mesh. Existing approaches fall short—prior fisheye Gaussian Splatting methods only rasterize up to 180° and cannot produce meshes, while Gaussian surface reconstruction methods assume pinhole training images.
We present GOFish, the first 3D Gaussian Splatting surface reconstruction method that trains directly on fisheye images, rasterizes FoVs beyond 180°, and extracts an aligned mesh from the same scene.
Key Contributions
GOFish introduces two technical components that together enable clean reconstruction across the entire FoV:
- Calibrated Densification recovers the surface across the full field of view, extending usable geometry beyond the 180° ceiling of prior work.
- Calibrated Mip Filtering and Wide-Angle Appearance Modeling keep pixel blur and lens falloff out of both the rendered scene and the extracted mesh.
Results
Across the 200° FIORD and 205° FullCircle benchmarks, GOFish improves PSNR over the best prior method by 2.28 dB and 1.97 dB, respectively. It also outperforms fisheye rendering baselines within their own FoVs and achieves the highest mesh F-score on FIORD and the main ScanNet++ set.
Why It Matters
Together, the Gaussian scene and mesh form a LiDAR-free, simulation-ready digital twin. The system renders in real time, ray traces the full FoV, and supplies collision geometry—making it directly applicable to robotics and autonomous navigation pipelines that depend on accurate, physically consistent environment models.
via ArXiv CV