A 3D head from a phone video
Fixed camera, slow head turn, and a pipeline that never sends a photo to the cloud.
I wanted my own head on this site, not a stock model. I also did not want my face uploaded to someone else's servers to get it.
What worked
The camera stays still on a tripod and I turn my head slowly, one full sweep. Then:
ffmpegpulls sharp frames out of the video.- A person-segmentation mask cuts me out of every frame, so the background cannot confuse the reconstruction.
- Apple's Object Capture builds a textured mesh from the masked frames, on my Mac mini.
- A Python script turns the mesh into the point cloud the website draws.
Without step 2 it fails: the room stays still while my head turns, so the software thinks the room is the object.
What is still wrong
The shape and the hair are closer to me than the previous scan. The lower face is not there yet. The nose, mouth and chin come out too faint and flat, and that is most of what makes a face look like a particular person.
The site still shows the old head until the new one actually looks like me. No sculpting allowed: it has to come from the scan.