Bumble Bee (Bombus terrestris)
Capture
_____________________________________
- Camera: SonyA7R5 camera (61 Mp) - 1/40 sec. - f16 - ISO 100 - JPG.
- Lens: Laowa 90 mm 1-2 x Macro at ~ 0.7 x
Scan
_____________________________________
- Scanned with: MICROscanR - a 3D focus-stacking robot (home build).
- Total viewpoints (VPs): 631 (optimally distributed along a spherical scan shape)
- Average stack-depth per VP: 31 frames, ranging from 22-38 (dynamically assigned per viewpoint).
- Total pictures: 19.666
- Scan time: 4h09m.
- Scan date: 28-09-2026.
Splat
______________________________________
- Total splats: 5.000.000
- image resolution used: 4096 px width (full res = 9504 x 6336 px)
- Software: Lichtfelt Studio
Lighting
______________________________________
- Lights: Object in diffuse light field between 2x side-beam round (28 cm) LED panels, 2 x 24 Watt
+ 1x Microscope LED beamer, with laser-cut PMMA diffuser - sand-blasted, fixed on front of lens.
- Background: black velvet.
Post processing
______________________________________
- Completely scripted using Helicon and Reality Scan (CLI) as external software - All steps progress-monitored in web based UI.
- Camera poses were transferred from a scan of a calibration object (doodle ball), taken from the exact same VPs without focus stacking.
- These poses were corrected for the focus stacking depth per viewpoint
- Then they were used as starting point for object camera alignment in Reality Capture.
- Missing cameras (that got rejected during the alignment - 40 in this case) were later calculated from neighbouring cameras and injected into the dataset (COLMAP format).
- I call this method "Hybrid Alignment Transfer with Injection".
- This was the first official test of this method. This Bumble Bee would not have aligned well without this method.
- Total post-processing time: 2h05m