EOGS++: Earth Observation Gaussian Splatting with Internal Camera Refinement and Direct Panchromatic Rendering

TL;DR

EOGS++ tackles robust 3D reconstruction from satellite imagery by replacing external preprocessing with an internally refined bundle adjustment based on optical flow, and by operating directly on high-resolution pan-chromatic data. It extends Gaussian Splatting to Earth observation with an affine camera model, per-camera color correction, shadow handling, and elevation-based rendering while incorporating opacity resetting and TSDF-based postprocessing to sharpen DSM outputs. Across the IARPA2016 and DFC2019 datasets, EOGS++ achieves state-of-the-art accuracy and efficiency, outperforming previous NeRF-like methods and the original EOGS, with a notable MAE improvement on buildings from $1.33$ m to $1.19$ m. These contributions reduce reliance on preprocessing steps, streamline camera pose refinement, and yield sharper, more reliable digital surface models for remote sensing applications.

Abstract

Recently, 3D Gaussian Splatting has been introduced as a compelling alternative to NeRF for Earth observation, offering com- petitive reconstruction quality with significantly reduced training times. In this work, we extend the Earth Observation Gaussian Splatting (EOGS) framework to propose EOGS++, a novel method tailored for satellite imagery that directly operates on raw high-resolution panchromatic data without requiring external preprocessing. Furthermore, leveraging optical flow techniques we embed bundle adjustment directly within the training process, avoiding reliance on external optimization tools while improving camera pose estimation. We also introduce several improvements to the original implementation, including early stopping and TSDF post-processing, all contributing to sharper reconstructions and better geometric accuracy. Experiments on the IARPA 2016 and DFC2019 datasets demonstrate that EOGS++ achieves state-of-the-art performance in terms of reconstruction quality and effi- ciency, outperforming the original EOGS method and other NeRF-based methods while maintaining the computational advantages of Gaussian Splatting. Our model demonstrates an improvement from 1.33 to 1.19 mean MAE errors on buildings compared to the original EOGS models

Figures (6)

• Figure 1: Qualitative result of the EOGS++ pipeline.
• Figure 2: Schematic overview of the proposed training pipeline. From the 3D Gaussian primitives, an image is rendered. The rendered image is then aligned with the training observation using an optical flow algorithm, after which the model is trained accordingly.
• Figure 3: Zoomed-in DSM comparison for respectively the IARPA 001, IARPA 002, and JAX 214 with and without opacity reset. The reset operation helps eliminate Gaussian floaters.
• Figure 4: Qualitative comparison of different methods for handling errors in the camera pointing (cf.\ref{['tab:RPC_quantitative']}) on IARPA 002 and IARPA 003 scenes.
• Figure 5: Rendered images on IARPA 001 using different B.A. strategies. Notice how the images become sharper.