CultCap
CultCap · Creator and developer · 2026
- Gaussian splatting
- gsplat
- PyTorch
- COLMAP
- Kotlin
- Camera2
- Python
- Cloudflare
A minute of walking around an object with your phone. Back comes a 3D model you can see from any angle.
CultCap is a phone capture app and a Gaussian splat pipeline for objects: a sculpture in a regional gallery, an artefact in a historical society’s back room, a product for a shopfront. No markers, no rig, no studio lighting, no manual clean-up. The capture uploads, the pipeline removes the room, and what comes back is something you can put on a website: a turntable, a video, an image from any angle.
From the gallery floor
A piece behind glass, and a long, thin wall piece (Puff, John Turier, 2004) captured from one side only. The app tracks which angles the capture covered, and the camera path stays inside them, so the model is never shown from somewhere it wasn’t photographed.
Photograph vs reconstruction
Drag the handle: the left half is the photograph, the right half is the reconstruction rendered from the same camera position, with the room removed. This photograph was held out of training, so the model never saw it.
How it’s built
- Capture app: native Android (Kotlin, Camera2). Full-resolution stills with per-frame exposure, focus and motion-sensor metadata, because a splat reproduces every quirk of the camera.
- Pipeline: structure from motion on the original photographs, never resampled; automatic background removal; frames that disagree with the consensus object are dropped; gsplat training with lens distortion handled at the loss.
- Measured, not guessed: every change is scored on held-out photographs before it’s adopted, and dropped if it doesn’t earn its place.
Writing
- Why I’m doing this
- A day at the Newcastle Art Gallery
- Free focus and camera intrinsics
- All CultCap posts
Curators, artists, anyone with an object worth capturing: john@cultcap.app.
