Egocentric Marquetry and Inlay Data

Egocentric marquetry and inlay data is first-person video of a person doing marquetry, recorded from the contributor’s own point of view with synchronized sensor streams alongside. The camera sits on the head, close to what a robot’s onboard camera would see, so the frames show the hands on the veneer pieces and the inlay taking shape. On this site it is a custom collection: you name the task, the materials and the setting, and capture is run to a written skill spec.

The data is human capture by paid, consented contributors. It is not robot or teleoperation data, and it is not a fixed corpus you download. People reach it as a veneer marquetry dataset, first-person wood inlay video or custom woodcraft data collection; the work is the same. It is a different subject from egocentric veneer lamination and pressing data, which lays new veneer sheets flat on a substrate.

A published robot foundation model paper, GR00T N1, linked in the sources below, describes a data pyramid with web data and human video at the base, synthetic data in the middle and real robot trajectories at the top. First-person human video of decorative inlay work sits in that base layer. Human capture does not replace robot data, and this page does not claim that it does.

The site’s commercial task catalog lists one row for this subject. "Marquetry" sits in the Woodcraft & Carving Studio environment, described as "Cut veneer pieces to pattern and glue them into a decorative inlay on a panel." It is listed as hard, captured with a phone (head-mounted) or multiview rig, and marked available on request. The first table below reproduces the row exactly as the catalog states it.

"Available on request" is not "ready to deploy", which is how the neighbouring "Veneer Lamination & Pressing" row is listed. The site does not describe a pattern, piece count, panel material, veneer type, adhesive or cutting tool for the row, so none is stated here.

The catalog states the capture setup for the Marquetry row as "Phone (head-mounted) / multiview rig". This page reports that setup and does not describe the row as stereo capture. The site publishes a separate page for stereo woodcraft and carving data, which the custom collection hub describes as covering marquetry among other studio crafts; if you want stereo, raise it in the spec.

Across Firsthand programs the wider stack is ten streams: a 4K head camera, two wrist cameras, depth, IMU, 3D hand and body pose, instance segmentation and action labels, hardware-aligned under 2 ms on one PTP clock domain. The wrist cameras give a closer view of the hands than the head camera alone. The site states no camera count, resolution or frame rate specific to marquetry, and force and tactile readings are not among the streams.

Action segments carry start_ns, end_ns, verb and noun, so the catalog row becomes a run of labeled steps rather than one long clip. The other standard fields are hand_joints and hand_visible for 3D hand pose, contact_events for the moment the hand meets an object, objects, and instance_masks stored at 848×480 to line up with depth. Hand pose covers both hands with a per-joint visibility flag for occluded fingers.

The site does not list labels for the inlay itself: no pattern, piece count, fit quality, gap width, glue coverage or flatness label, no task-completion score, and no contact force or pressure reading. It also does not say whether a veneer piece or the panel counts as the item in a contact event, so the spec settles that.

Fingers disappear behind objects in manipulation, which is why hand pose carries a visibility flag. The site says depth degrades on wet and reflective surfaces and handles that with per-frame confidence maps; it does not say how depth behaves on a glued inlay or on small pieces lying flat against a panel.

An episode is rejected for a sync breach beyond the ceiling for more than three consecutive frames, calibration drift beyond tolerance, the manipulated object out of frame for more than 15 percent of the episode, more than 35 percent invalid depth pixels in the manipulation region, an undocumented bystander, or a mismatch against the named skill spec. A human reviewer then checks each episode. For marquetry, the spec decides what the manipulated object is. No mask accuracy figure for veneer pieces or panels is stated.

Every contributor signs a written release before capture, and each delivered episode links to that record through consent artefact IDs. At ingest, faces, plates and screens are blurred irreversibly, speech is scrubbed of names, addresses and card numbers, and a human reviewer confirms that no un-consented person is identifiable in any frame. Footage that cannot be de-identified is discarded, and the anonymization record is delivered with the data.

The site states no contributor safety protocol and no trade certification for the Marquetry row, and lists capture at your own facility as not offered. None of this is legal advice, and the executed agreement controls.

From signed spec to first validated batch, the timeline is a median of 48 hours where coverage is deep and three to four weeks where it is thin and contributors need to be recruited first. The Marquetry row is marked available on request, and the site does not say which applies to the studio, so ask when you request the quote. After the first batch, on-demand top-ups run against the same accept criteria. Collections are billed per validated hour, and the site publishes no price list, minimum order or batch size for this task.

Delivery is in RLDS, LeRobot, WebDataset, HDF5, zarr or Rerun .rrd, or raw streams with metadata, with readable converter source. The license is perpetual, irrevocable and worldwide, buyer-owned and non-exclusive by default, with exclusivity available as a separate term.

Describe the patterns, the panels, the setting and the variation or failure cases your pipeline is missing. That becomes a written skill spec, quoted for reach, timeline and price before any capture begins, and most programs run a Spec Pilot first, a small, fast first batch that proves the spec and accept criteria on real footage. Name each step rather than a broad activity, name the setting and capture tier, and name where an episode starts and stops. A published sample recording is on the sample data outputs page if you want to see the streams first.