Egocentric Gem Setting and Engraving Data

It is first-person video of a person doing gem setting or engraving at a jeweler bench, 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 while it handles a small part, so the frames show the hands on the stone, the setting and the tool, rather than a distant view of someone at a bench. On this site it is a custom collection: you name the task, the setting and the variation, and capture is run to a written skill spec.

The data is human capture. A paid, consented contributor wears the rig and does the task. It is not robot or teleoperation data, and it is not a fixed corpus you download. People reach this topic through close cousins: jewelry robot training data, an egocentric gem setting video set, a hand engraving under magnification dataset, or custom jewelry studio data collection. They describe one request from different angles, and the work is the same: write down the task, collect it on demand, and receive validated episodes.

The two catalog rows are about setting a stone and engraving metal. Nothing on this page states that the video carries a stone label, a setting-quality score, an engraving-quality label or a measurement.

Two. Both sit in the Jewelry & Metalsmithing Studio environment of the Creative Workshops & Crafts sector, whose listed examples are goldsmith bench, gem setting, wire wrapping, soldering, engraving and polishing. "Gem setting" is described as: "Seat a stone in its setting and push or bead the prongs over it with fine tools." "Engraving" is described as: "Cut letters and patterns into metal with gravers or a rotary tool under magnification." Both are rated Hard, both list a head-mounted phone or a multiview rig as the capture setup, and both are listed as Available on request.

The site says nothing further about what Available on request means in volume, hours or delivered footage, so this page claims no completed collection and no delivered hours, and what applies to a given order is confirmed when you request the quote. This page names only the objects, tools and steps that appear in the rows: a stone, its setting, prongs, fine tools, letters and patterns, metal, gravers, a rotary tool and magnification. The first table below reproduces both rows as the catalog states them.

The egocentric jewelry and metalsmithing page covers all five rows of the Jewelry & Metalsmithing Studio environment: Engraving, Gem setting, Polishing, Soldering and Wire wrapping. On that page, gem setting and engraving appear as two rows among five, each with a short section, and none of the five is the page’s main topic.

This page is the one place that treats gem setting and engraving as the subject. It does not cover Soldering, Wire wrapping or Polishing, which are different rows with different steps. The second table below sets the pages side by side.

Both tasks are done close, with small tools, and the hands are present throughout. The site states that egocentric capture sits close to what a robot’s onboard camera would see, and that 3D hand pose carries what grip and finger changes look like when plain video cannot resolve them. Human capture does not replace robot data, and this page does not claim that it does.

A custom collection also lets you aim at the step your model is missing, such as the prong work after the stone is seated, rather than hoping a broad collection happens to contain it. The site lists no off-the-shelf gem setting or engraving dataset, so this work is custom collection, and this page does not claim that any public corpus lacks jewelry bench data.

Each attempt is delivered as one episode, and these tasks need a decision up front about what one attempt is: one stone seated and its prongs worked, one letter or pattern cut, or a longer run of pieces. The site does not set that length, so the skill spec does. Inside the episode, the synchronized streams run from start to end, and action segments divide it into labeled spans.

An episode might open with a stone being placed in its setting and close with the last prong pushed or beaded over it, as the catalog row words the work. That is an illustration of how a spec can cut an episode, not a stated duration. How many pieces are covered is set in the spec. The site states that Firsthand does not capture at a customer’s own facility, so the bench or studio used is agreed in the skill spec.

A gem setting or engraving collection uses the same capture stack as every Firsthand program. The instrumented tier is a head-mounted stereo rig that records first-person RGB, with a metric depth stream recovered from the calibrated stereo rig rather than from a standalone depth sensor, and 6-DoF IMU data at 200 Hz from head- and wrist-mounted sensors. 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.

Every stream is hardware-triggered onto one PTP clock domain, so depth, motion and hand pose line up with the same video frame rather than being approximated afterwards. The rig is calibrated twice per session, before capture starts and after it ends, and an episode is accepted only if RMS reprojection stays under 0.28 px. The catalog lists a head-mounted phone or multiview rig for both rows; a multiview rig adds synchronized third-person angles when the spec asks for them. Which tier applies to your program is set in the skill spec.

The site does not state a camera count, resolution or frame rate specific to these two rows. Whether a magnified view is part of the recording is not stated, and neither is the magnification equipment or level, so raise it in the spec if your policy needs it. Force and tactile readings are not among the streams, and no robot or teleoperation data is captured.

Inside each episode, action segments divide the recording into start- and end-timestamped spans, each carrying start_ns, end_ns, verb and noun. For these tasks a segment can be one step as written in the skill spec, so a catalog row becomes a run of labeled steps such as seat and push or bead, or cut, 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 the depth resolution of 848×480 so they line up with the depth stream.

Hand pose is delivered for both hands with a per-joint visibility flag for occluded fingers, 21 joints per hand in 3D. The site describes contact events as the moment the hand meets an item, and it does not say whether a stone, a prong or a graver counts as the item, so the spec is the place to settle what a contact event means for your task.

The site does not list labels for the stone, the setting or the engraving. There is no stated label for stone type or size, setting style, how well a stone sits, cut depth, lettering or pattern, quality, tolerance or task-completion score, and no contact force reading. The catalog description names the work the contributor performs, not a delivered measurement.

Several limits the site already states apply directly to bench jewelry work, and several it leaves open. Each stated limit below appears elsewhere on the site rather than being introduced here.

Hand occlusion is the first. Fingers disappear behind objects in manipulation, which is why 3D hand pose carries a per-joint visibility flag, and the contact events and per-frame masks sit alongside it. Motion blur and hands entering the frame from below are part of what egocentric video puts into the training distribution. Depth is the third. 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 metal or a stone in particular.

The fourth is the set of machine-checked reject criteria. 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 against the task spec. For these tasks, the spec decides what the manipulated object is: the stone, the setting, the prongs or the piece being engraved.

Also not stated: how often a small piece leaves the view at a bench, a blur or sharpness rule, a mask accuracy figure for stones or settings, and any description of how a part fully covered by the hand is handled. If your policy depends on any of these, say so at the start.

Validation runs against signed accept criteria written into the spec before capture. Each batch goes through per-batch QA and a reject log ships with it. Only footage that clears the criteria is billed as a validated hour, and rejected items are recaptured at our cost. Many teams start with a Spec Pilot, a small, fast first batch that proves the spec and the accept criteria on real footage before volume scales.

Every contributor signs a written release before capture, and each delivered episode links to that record through consent artefact IDs in the batch data card. At ingest, automated detection finds faces, plates and screens and blurs them irreversibly, and speech is scrubbed of names, addresses and card numbers while ambient audio is kept. A human reviewer confirms that no un-consented person is identifiable in any frame, and footage that cannot be de-identified is discarded. The anonymization record is delivered with the data.

On safety and quality, the site states no contributor safety protocol, no guidance on protective equipment, no cycle-time, throughput, tolerance or defect-rate figure, and no metal, stone-grade, quality-system or certification claim for these rows, and this page does not describe one. None of this is legal advice, and the executed agreement controls.

The site offers a tactile glove tier as a pilot-ready option. The gloves track the hands, the glove signal is time-synced to a head-mounted stereo camera by hardware timestamps and an IMU, each glove is calibrated every session, and the pipeline checks signal dropout and sync drift.

The site does not state that the glove tier has been used on either row, and it does not state whether gloves fit fine bench work. Treat force and tactile data as Not stated for these rows unless a spec adds the glove tier and agrees the fit with you. The tactile glove page covers the tier in full.

From signed spec to first validated batch, the timeline is a median of 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. Speed comes from a warm pipeline, not from cutting consent, review or anonymization. Both rows are marked Available on request, and the site does not say whether the jewelry studio environment counts as deep or thin coverage, so ask which timeline applies when you request the quote. After the first batch, on-demand top-ups run against the same accept criteria, so you can add more pieces or steps without rewriting the spec. The site does not state a turnaround for those later batches, so plan them in your quote.

This is custom collection at scale through a network of thousands of capture providers, for a consistent, recurring data flow to labs. Delivery is in the format your stack already reads: RLDS, LeRobot, WebDataset, HDF5, zarr or Rerun .rrd, or raw streams with metadata. Readable converter source ships alongside the data so the schema can be retargeted to your own loader. Collections are billed per validated hour.

The license is perpetual, irrevocable and worldwide, with the data buyer-owned and non-exclusive by default. Training, evaluating, fine-tuning and deploying models, including commercial ones, is permitted, and exclusivity is available as a separate term. The executed agreement controls. The site does not publish a price list, a minimum order or a batch size for these tasks, so those are set in your quote.

Where this site does not state something, this page says "not stated" rather than filling the gap. The fourth table below separates what is delivered from what is not stated. There is no listed label for stone, setting or engraving quality, no force or tactile stream, no metal, stone or tool specification beyond the words in the two rows, no magnification level, and no throughput, cycle-time or certification claim. Nothing here claims a training outcome or a model improvement, and no delivered bench footage or hour count is claimed.

The tradeoff to plan for is the embodiment gap between a human hand holding a small piece and a robot platform doing the same, which retargeting or pairing with robot data has to close. Counts of contributors, hours and pieces are set in your skill spec.

Describe the two tasks and where each starts and ends, 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. The fifth table below maps the parts of a skill spec onto a request for these rows. A published sample recording is on the sample data outputs page if you want to see the streams before you write a spec.

Three choices matter most for these tasks. Name each step rather than a broad activity such as "jewelry work", because that is what lets each step become one action segment. Name the setting and the capture tier, since the catalog lists a head-mounted phone or multiview rig for both rows. And name where the episode starts and stops: one stone, one cut or a run of pieces, since the site does not set an episode length.