Factory pDigger · Research

One frame in. Scene data out.

Two camera frames from a working production cell, each taken apart in seven stages. Orbit them and switch layers: surfaces and depth, the worker as measured points, a skeleton and a body model, and every stone by kind — with the holes between them.

Loading the scene…

Depth ×0

Stage 01 of 07

Frame

It starts with a single frame from one ordinary camera — no stereo rig, no depth sensor, no second view. We don’t publish the photo: what you see is where each pixel lands.

Customer value

Works where people already inspect: one ordinary camera over any conveyor, tray or pallet line with a human inspector — no stereo rig, depth sensor or retooling.
Every inspection becomes data: what is on the tray, of which kind, in which layer, and which slots are empty.
The inspector’s hands are part of the record: which item was touched, lifted or set down, and where.
Experienced judgement becomes measurable — something you can audit, compare across shifts and teach from.
People are kept as geometry only: body model and skeleton, never a face or an identity.
Each frame ships as a hashed scene package, ready for your quality, traceability or analytics systems.

Scene package

What’s in the scene package

Each sample ships as a small scene package: three binary layers and one description, which is all this page loads.

Depth: One relative depth value per point, stored as 16 bits: about 1 at the work surface, smaller towards the camera. A ratio, not millimetres.
Surfaces: The direction each point’s surface faces.
Labels: What each point is: the tray, one of the stones (numbered at random, as reviewed: every visible stone, none on the base plate), the worker’s body, the left or the right hand, a hole — or no data, where sample B’s frame was rotated upright and its corners stayed empty.
Description: The generic camera that lifts the points into 3D and how to decode each layer; the fitted tray surface that levels the base; one size per stone kind, and for every stone its kind, layer, footprint and centre; the holes; the worker’s skeleton — every joint’s position and whether it was measured on the visible body or inferred from the fitted rig — with its bones and the body model’s capsules; counts; honesty flags; and a SHA-256 hash for every file.
Simplified: Real frames from a working production cell, cropped to the work area and downsampled ×2; a generic camera instead of the real calibration; relative scale, not millimetres. Hidden surfaces are never invented; hidden joints come from the fitted body rig and are marked.
Left out: The photo, the camera’s real calibration, names, timestamps and the pipeline’s model identifiers. People appear only as geometry.