Copy & Paste the World with Gaussian Splatting - Scott Leaman - NDC Oslo 2026

Copy & Paste the World with Gaussian Splatting - Scott Leaman - NDC Oslo 2026

by Scott Leaman · View the channel on PSI

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQVComputer vision
🎙 Scott Leaman 👥 230K 📅 October 14, 2026 ⏱ 34 min 👁 15 📄 expert opinion 🧭 2026-10-14
Available in: English (current) Français

Keywords

Gaussian Splatting3D captureLiDARDigital TwinVR

Summary

Scott Leaman, spatial engineering lead at Sopra Steria Norway, presents a practical overview of Gaussian Splatting for creating and displaying 3D scenes of the real world. He begins by contrasting traditional capture methods (LiDAR, depth cameras, photogrammetry) and display techniques (point clouds, voxels, ellipsoids, meshes), highlighting trade-offs between accuracy and human readability. He illustrates the value of 3D capture through a detailed case study of the French railway SNCF, which uses LiDAR-on-train systems to create digital twins for infrastructure planning and collision testing. He then explains Gaussian Splatting itself: how scenes are built from millions of colored ellipsoids, how spherical harmonics capture view-dependent effects, and how files can be stored (PLY, SPZ). He demonstrates multiple captures made with smartphones, 360 cameras, drones, and VR headsets, and shows how splats can be edited (e.g., SuperSplat, Blender, Unity) and rendered in browsers via three.js or game engines. He covers generative tools like Marble from World Labs, which synthesizes splats from text prompts. Applications discussed include real estate, heavy industry, defense, construction, museums, and film special effects (e.g., Superman). Finally, he gives a practical guide to getting started with free tools like Niantic Scanniverse and paid ones like Varjo Teleport, noting trade-offs in data ownership and quality.

204 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk delivers substantial practical value by demystifying Gaussian Splatting and providing concrete, replicable workflows. Leaman supports his claims with real examples from his own projects, including captures of Vigeland Park, Oslo rooftops, and a statue, demonstrating the technology’s capabilities and limitations. He also offers a clear cost-benefit analysis of different capture methods, making a strong case for why Gaussian Splatting fills a niche between costly LiDAR and lower-quality photogrammetry. His argumentation is well-structured, building from basic concepts to advanced applications, and he does not overhype the technology—he candidly notes where splats fail (e.g., close-up blur, data size, limited editing maturity). The inclusion of a live example with the SNCF railway digital twin provides a compelling, enterprise-grade validation of the approach. The talk is convincing and actionable, with a clear ‘how to try it yourself’ section at the end.

Scientific Rigor, Source Quality, Title Accuracy

The talk does not cite formal scientific papers, but it references several well-known tools and platforms (NVIDIA Omniverse, CloudXR, Apple Vision Pro, three.js, SuperSplat, Marble, Niantic Scanniverse, Varjo) which are verifiable. The speaker’s 20 years as a developer and 11 years in XR lend credibility, and the examples shown are from real deployments. However, the lack of external sources or data comparisons weakens the rigor slightly. The title is an apt metaphor for the concept of capturing and relocating real-world scenes, and the content matches it closely, covering both capture and display. The description provides only conference links, which are used for sources_citees. Overall, the talk is reliable for practical insights but not as a source of academic evidence.

273 words

Title / Content Match

The title accurately reflects the content, which focuses on capturing and reproducing real-world scenes using Gaussian Splatting.

Quality & Reliability

7/10

The talk is based on the speaker's extensive practical experience in spatial engineering and provides numerous real-world examples, but it does not cite formal scientific sources or peer-reviewed literature.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • NDC Conferences — Official conference website where this talk was presented (NDC Oslo).
  • NDC Oslo — Specific conference page for NDC Oslo, the event where the talk was recorded.

External References

Contribution & Novelties

The talk provides a comprehensive, practitioner’s perspective on Gaussian Splatting, bridging the gap between academic descriptions and hands-on use. It uniquely combines an overview of capture methods, detailed technical explanations, and a pragmatic guide to tools and workflows, all grounded in real project examples. The speaker’s emphasis on enterprise applications (railway, real estate, heavy industry) adds value beyond typical hobbyist demos. For further exploration, the following concepts and references are directly related to the core ideas of the talk:

Pour aller plus loin :

  • 3D Gaussian Splatting (Wikipedia) — Overview of the technique, its mathematical basis, and applications.
  • Neural Radiance Fields (NeRF) — A related depth-based 3D reconstruction method that predates and complements Gaussian Splatting.
  • Digital Twin — Concept of creating virtual replicas of physical systems, as applied in the SNCF case.
  • Photogrammetry — Classical technique for 3D reconstruction from photos, contrasted with Gaussian Splatting in the talk.

148 words

Radar Profile

The radar profile shows a balanced distribution across all dimensions, with slightly higher scores in quantity of information and technical level, while quality and reliability are slightly lower, reflecting the practical but non-academic nature of the talk. The high technical level is supported by detailed explanations and demos, but the lack of citations and peer-reviewed backing keeps reliability moderate.

Reliability 7/10