360 Scanner

Scan

How to scan

Sit on a swivel chair (or stand) with your head and shoulders in view, good light, plain background helps.
Face the camera and hold still for a second.
Turn slowly, about 20° a second, in either direction. Two full turns, ending facing the camera.
The head is rebuilt and shown in 3D.

turn 1turn 2

Quality

Resolution

Mirror the camera view

Result

Colours

Turntable

Above: the captured photo nearest to the 3D view's angle.

Above: the baked texture, unrolled around the turning axis (front in the middle, the person's left to the right, crown at the top).

How it works

MediaPipe's pose model tracks how far you have turned from your shoulders, ears and nose, and every time your face looks straight at the camera the count is pinned to 0°, 360° or 720°. Its multiclass segmenter cuts out hair, face and neck (head) or the whole person (body) in every frame; a body is centred on the pose model's hip and shoulder midline. The steadiest frame per 2° of each turn is kept, so every angle is seen twice. Space carving keeps the voxels that (almost) every silhouette agrees on and surface nets turn them into a mesh. The default view is image-based: the two full-resolution photos taken nearest the viewing angle are projected onto that shape and cross-faded, which stays photographic even where the shape is rough; a baked cylindrical texture fills what neither photo sees.

Limits: silhouettes cannot see hollows (eye sockets, armpits and the gap between the legs fill in), so keep arms a little away from the body. Head size assumes a 63 mm eye distance, body size comes from MediaPipe's metric landmarks; the camera is treated as orthographic. Nothing leaves the browser.