Research map / 3D vision / Gaussian splatting
RPC-GS: Gaussian Splatting with native RPC Rendering for Satellite Imagery
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces RPC-GS, a Gaussian Splatting renderer that projects Gaussian means and covariances natively through the RPC model for satellite imagery, using a geo-coordinate transform chain, Jacobian covariance projection, and metric ray-based depth.
Problem
Existing satellite Gaussian Splatting methods approximate RPC camera models with perspective or affine cameras, introducing systematic geometric errors in projection and covariance splatting that degrade reconstructed geometry.
Approach
Embed the RPC model natively in a Scene→ENU→ECEF→Geodetic→Image transformation chain, propagate covariances with a Jacobian chain (using a solver-independent inverse ECEF→geodetic Jacobian obtained from the forward map), and define a metric ray-based depth for front-to-back compositing.
Key results
- On DFC2019, RPC-GS reduces mean altitude MAE from 3.04 m (perspective) to 2.14 m (29.6%) and from 5.91 m (affine) to 2.14 m (63.8%).
- On IARPA2016, mean altitude MAE improves by 9.9% over perspective and 37.9% over affine (2.10 m vs 2.33 m and 3.38 m).
- RPC-GS has lower PSNR than perspective and affine approximations (30.06 vs 33.88 and 34.95 on DFC2019).
Surprising findings
- Despite better geometry, RPC-GS yields lower PSNR than the approximation baselines, a geometry-appearance trade-off the authors attribute to prior work.
Limitations
- First-order Jacobian propagation may be insufficient for the nonlinear ECEF→geodetic and RPC projections.
- RPC projection and splatting implemented in Python, adding computational overhead.
- Shadow modeling and regularization omitted, so absolute performance is not competitive with full satellite pipelines; no error bars reported.
Experimental setup
Datasets: DFC2019 IARPA2016
Benchmarks: DFC2019 IARPA2016
Metrics: altitude MAE (m) PSNR
Baselines: 3DGS perspective approximation (SatelliteSfM) EOGS affine approximation
Builds on: 3D Gaussian Splatting EOGS SatNeRF WildGaussians SatelliteSfM
Closest papers by meaning
- Gaussian Splatting for Efficient Satellite Image Photogrammetry (CVPR 2025 · similarity 0.91)
- Superpoint Gaussian Splatting for Real-Time High-Fidelity Dynamic Scene Reconstruction (ICML 2024 · similarity 0.89)
- IBGS: Image-Based Gaussian Splatting (NeurIPS 2025 · similarity 0.88)
- 3D Convex Splatting: Radiance Field Rendering with 3D Smooth Convexes (CVPR 2025 · similarity 0.88)
- Horizon-GS: Unified 3D Gaussian Splatting for Large-Scale Aerial-to-Ground Scenes (CVPR 2025 · similarity 0.88)
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.