Gaussian splatting produces the most photorealistic 3D scenes the industry has seen, and that has been enough to generate serious interest and serious skepticism. The open question is whether splats can carry measurement, analysis and deliverables, or whether they remain a presentation format.
This session takes the question seriously from four directions. Speakers cover bringing georeferenced splats into CAD so engineers can work with them alongside lidar, a side by side comparison of splats against meshes, DSMs, point clouds and object-based models across realism, accuracy, editability, file size and analytics, the case that advances in dense image matching already deliver what production workflows need without changing formats, and the pairing of splatting with SLAM for mobile capture as national 3D mapping mandates take hold. Expect disagreement. Attendees will leave with a decision framework rather than a verdict.
This session will include the following presentations:
Comparing Gaussian Splatting, Meshes, DSMs, Point Clouds, and Object-Based Models
Presented by Arkadiusz Szadkowski, Esri
From Gaussian splats to meshes, DSMs, point clouds, and object-based 3D models, this session explores how each 3D representation performs across realism, accuracy, editability, file size, and analytics. We anchor the comparison in practical GIS workflows, showing where each format excels today and where it falls short. Attendees will leave with a clear decision framework for choosing the right 3D form for mapping, inspection, planning, and communication.
Does Geospatial Production Need Gaussian Splatting? Advances in Reconstruction for Next-Generation Mapping Workflows
Presented by Bernhard Schachinger, Vexcel Imaging GmbH
This presentation explores how advances in image-based reconstruction continue to drive meaningful improvements across established photogrammetric workflows. We will examine recent developments in dense image matching and reconstruction technology that significantly improve the quality and reliability of derived elevation data while maintaining compatibility with existing production environments.
Attendees will learn how modern reconstruction approaches reduce noise in photogrammetric surface models, improve feature definition, and deliver more robust outputs even under less-than-ideal imaging conditions. The presentation will demonstrate why reconstruction quality remains one of the most important factors influencing downstream products such as DSMs, DTMs, orthomosaics, and 3D city models.
The session will also discuss the broader industry direction. While Gaussian Splatting is gaining momentum for specific use cases, geospatial production organizations must balance innovation with scalability, operational efficiency, data quality, and long-term workflow sustainability. We will explore how continued innovation within established photogrammetric frameworks can unlock substantial performance and quality gains without requiring organizations to replace mature processing ecosystems.
Through real-world examples and technology insights, this presentation offers a practitioner’s perspective on the future of geospatial reconstruction and why advances in dense image matching remain highly relevant in an era increasingly shaped by AI-driven methods and emerging 3D representations.
Gaussian-SLAM: The Next Generation Reality-Capture Using Integrated Gaussian Splatting and SLAM
Presented by Mohamed Mostafa and Vi Huynh, Trimble Applanix
The rapid evolution of Simultaneous Localization and Mapping (SLAM) and 3D Gaussian Splatting is actively transforming the geospatial industry. Today, professional mappers routinely leverage multi-sensor data streams, utilizing these technologies independently to generate photorealistic 3D city models for diverse applications. This technological leap has catalyzed a global shift toward standardized 3D mapping deliverables. Nations such as Germany, Italy, France, the UK, and most recently Singapore, have established national 3D mandates. Globally, this transition is also being driven at the regional and municipal levels by pioneering agencies like the US Departments of Transportation (California, New York, Utah, and Ohio) and forward-thinking jurisdictions including Ottawa (Canada), Victoria and New South Wales (Australia), Helsinki (Finland), Rotterdam (The Netherlands), and Seoul (South Korea).
To meet the accelerating demand for high-precision, visually immersive deliverables, this paper presents a novel workflow that tightly integrates these traditionally independent technologies. In this proposed approach, global SLAM is deployed to optimize the multi-sensor data streams, establishing rigorous geometric constraints that maximize the geospatial accuracy of the Gaussian Splatting algorithm. Concurrently, Gaussian Splatting is utilized to generate a high-fidelity, photorealistic final presentation of the 3D map product.
We demonstrate this integrated methodology using datasets that combine airborne surveys with land-based mobile mapping systems. By fusing multi-perspective LiDAR and camera data from both aerial and ground vantage points, this unified SLAM-Splatting workflow delivers an optimal environment for comprehensive, survey-grade 3D urban modeling.
