Dear Support Team,
I am building an end-to-end drone mapping pipeline for city-scale 3D Gaussian Splatting (3DGS). My ultimate goal is to publish the final models as 3D Tiles into ArcGIS Pro / Enterprise, which strictly requires accurate Real-World Georeferenced Coordinates and a Z-up orientation.
Currently, I am facing a critical issue where the trained model loses its geographical location and orientation during the pipeline transition.
My Current Workflow
- SfM / Alignment: RealityScan (Version 2.2.0.119430) → Exported to COLMAP text format.
- 3DGS Training & Export: LichtFeld Studio (LFC Studio).
The Problem
When I export the COLMAP files from RealityScan and import them into LichtFeld Studio, the software trains the model, but it does not recognize, bake, or preserve the project’s real-world geographic location. As a result, the final PLY or 3D Tiles do not land on their correct geographic coordinates in GIS environments.
I suspect I am either using the wrong COLMAP export presets in RealityScan, or the georeferencing metadata (EXIF/GPS) is being stripped/misaligned during the handoff to LichtFeld Studio.
My Questions
- RealityScan Export Presets: What are the best export settings or presets in RealityScan (v2.2.0) to ensure that camera GPS/GCP data is properly baked into the COLMAP text export?
- Georeferencing & Transformation: What is the best practice to bake the georeferencing (EPSG / WGS84) and Z-up rotation directly into the 3DGS model (including Spherical Harmonics rotation) within this specific pipeline so it perfectly aligns with real-world coordinates?
- LichtFeld Studio Compatibility: Is there a specific alignment step, transform matrix, or configuration file I need to apply during or after the COLMAP import in LichtFeld Studio to maintain the coordinate system for 3D Tiles export?
Thank you for your time and guidance. I look forward to your expert recommendations to help resolve this alignment bottleneck. Below is my presets for Colmap Export
