MrNeRF (@janusch_patas)
2025-06-24 | โค๏ธ 66 | ๐ 7
The camera meshes donโt blend correctly with the splats yet but it looks kind of beautiful! Red indicates test cameras; green are training cameras.
What would your ideal workflow look like? Typically, when coming from COLMAP, the scene isnโt perfectly aligned to the target coordinate system.
Hereโs what I imagine the ideal process might be:
- Load COLMAP data
- Manually align the scene to the appโs coordinate system
- Start training
Some features that would make this smoother I can think of:
- Clean up the point cloud before training
- Optionally pause training to remove floating artifacts
Anything else youโd add? Iโm trying to shape a user-friendly, intuitive workflow!
๐ Related
See similar notes in domain-vision-3d