A macro photographer captured a strawberry across 90 perspectives with 88 focus-stacked images each, trained into a Gaussian splat via slang-splat and COLMAP.
Key Takeaways
Capture rig: Nikon Z8, Laowa 180mm macro lens, f/7.1, ISO 100, 1/160s, LED light, bluescreen background.
90 camera positions x 88 focus-stacked frames per position = ~7,920 source images fed into reconstruction.
Training pipeline: COLMAP for camera registration, slang-splat (github.com/MichaelMoroz/slang-splat) for Gaussian splatting.
Result released CC BY 4.0 on SuperSplat; attribution appreciated but not required.
Hacker News Comment Review
Commenters noted Gaussian splats degrade visibly on close zoom, flagging a known resolution ceiling for macro-scale subjects that microscopy rigs might push further.
The macro setup prompted immediate interest in combining microscopy with Gaussian splatting, with discussion around whether better lighting and optics could achieve sub-millimeter fidelity.
The author shared full rig photos and noted a separate splat of their own studio exists on SuperSplat, described as “painterly” in quality, implying indoor multi-object scenes remain harder to reconstruct cleanly.
Notable Comments
@voidUpdate: “as soon as you zoom in, they really fall off a cliff” – identifies the core quality limit for macro splats.
@josh-wrale: proposes combining microscopy hardware with Gaussian splatting as a natural next step.
