RPC-GS: Gaussian Splatting with native RPC Rendering for Satellite Imagery

2026-06-04 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Geospatial AI · Depth: Expert, quick

Summary

RPC-GS is introduced as the first Gaussian Splatting framework designed for satellite imagery that natively integrates Rational Polynomial Camera (RPC) models. Unlike prior methods that rely on perspective or affine camera approximations, leading to geometric errors, RPC-GS directly projects Gaussian means and covariances through the RPC model during splatting. It achieves this by embedding the RPC model within a chain of geo-coordinate transformations, deriving a robust Jacobian-based covariance projection for nonlinear transformations, and integrating a metric ray-based depth formulation. Benchmarking shows RPC-GS consistently yields the lowest reconstruction error on leading satellite datasets, improving mean altitude error by 29.6% and 63.8% on DFC2019, and by 9.9% and 37.9% on IARPA2016 compared to perspective and affine approximations, respectively. The code is released to support future research.

Key takeaway

For Computer Vision Engineers and AI Scientists developing 3D reconstruction from satellite imagery, RPC-GS offers a significant advancement in geometric accuracy. By natively supporting Rational Polynomial Camera models, it eliminates the errors introduced by traditional perspective or affine approximations. You should consider integrating RPC-GS into your workflows to achieve superior reconstruction fidelity, especially for applications demanding precise geospatial modeling.

Key insights

RPC-GS improves satellite imagery 3D reconstruction by natively integrating Rational Polynomial Camera models into Gaussian Splatting, avoiding geometric approximations.

Principles

• Native RPC integration avoids geometric errors.
• Jacobian-based projection handles nonlinear transformations.
• Metric ray-based depth addresses RPC's depth absence.

Method

RPC-GS projects Gaussian means and covariances directly through RPC models, embedding RPC in geo-coordinate transformations, using a Jacobian-based covariance projection and metric ray-based depth formulation.

In practice

• Use RPC-GS for higher fidelity satellite 3D models.
• Benchmark RPC against perspective/affine models.
• Explore released code for satellite imaging research.

Topics

• Gaussian Splatting
• Satellite Imagery
• Rational Polynomial Camera
• 3D Reconstruction
• Computer Vision
• Geometric Modeling

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• RPC-GS: Gaussian Splatting with native RPC Rendering for Satellite Imagery — Native RPC integration in Gaussian Splatting significantly improves satellite imagery 3D reconstruction accuracy.
• NG-GS: NeRF-Guided 3D Gaussian Splatting Segmentation — NG-GS enhances 3D Gaussian Splatting segmentation by integrating NeRF-guided continuous feature fields to resolve boundary artifacts.
• Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting — LGTM enables 4K novel view synthesis by decoupling geometry from resolution using textured Gaussian primitives.
• 3D Skew-Normal Splatting — Skew-Normal Splatting improves 3D scene representation by using skew-normal distributions for better modeling of asymmetric structures.
• GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens — GlobalSplat uses a global latent scene representation for efficient, compact, and consistent 3D Gaussian Splatting.
• AMD-Powered 3D Gaussian Splatting for Autonomous Driving Scenes — Street Gaussians on AMD GPUs enables dynamic 3D scene reconstruction and editing for autonomous driving.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.