Egocentric Jewelry Making and Metalsmithing Video Data

Egocentric jewelry and metalsmithing data is real-world, first-person video of work at a jeweler bench. It is recorded from the worker viewpoint, with depth, 3D hand pose, contact events and frame-accurate action segments aligned on one hardware clock. The tasks it covers are the ones the site states in its task catalog: engraving, gem setting, soldering, wire wrapping and polishing.

This is fine craft work, where the hands and a small piece of metal do most of the job. The tool use page covers hand-tool and tool-mediated manipulation in general, the dexterous manipulation page covers fine hand work across many settings, and the pottery page covers clay. This page covers the jewelry bench, with the five tasks the catalog actually lists.

The site states jewelry work in its task catalog, in a Jewelry & Metalsmithing Studio environment under the Creative Workshops & Crafts sector. The environment lists five rows: Engraving, Gem setting, Polishing, Soldering and Wire wrapping. Its examples are goldsmith bench, gem setting, wire wrapping, soldering, engraving and polishing.

Every row uses the same capture mode, a head-mounted phone or a multiview rig, and every row is listed as Available on request. Three rows are rated Hard and two are rated Medium.

The catalog row reads "Cut letters and patterns into metal with gravers or a rotary tool under magnification." It is the stated cutting task, and it names two ways of doing it, gravers or a rotary tool.

The row is rated Hard. The metal, the lettering or pattern, the magnification equipment and the piece size are Not stated. Whether the magnified view is part of the recording is also Not stated, so raise it in the spec if your policy needs it.

The catalog row reads "Seat a stone in its setting and push or bead the prongs over it with fine tools." It covers placing a stone in its setting and then closing the prongs over it.

The row is rated Hard. Stone types, stone sizes, setting styles and the fine tools used are Not stated, and so is any success or quality label for how well a stone sits. Those are defined in the spec.

The catalog row reads "Flux the joint, heat with a torch and flow solder to join metal pieces." It runs from fluxing the joint through heating to flowing the solder, so one row covers the whole joining sequence at the bench.

The row is rated Hard. The site names a torch and no more. Torch type, solder type, temperatures and safety practice are Not stated, so this page makes no claim about them. They are set in the written spec.

The catalog row reads "Bend and wrap wire around stones and beads with pliers to form jewellery." It covers bending and wrapping wire by hand and with pliers, around stones and beads.

The row is rated Medium. Wire gauges, stone and bead types and the finished designs are Not stated, so the spec sets which pieces the session makes.

The catalog row reads "Polish metal pieces on the wheel with compounds and finish by hand." It covers machine polishing on a wheel and then a hand finish.

The row is rated Medium. Wheel models, compound types and the finish a piece is meant to reach are Not stated, and the site states no outcome label for how a piece looks afterwards.

A Stained glass assembly row in the Glass & Mosaic Studio environment mentions soldering joints, the electronics page covers component soldering, and a welding page covers metal fabrication. They sit in other environments with other tasks, so this page lists them as neighbours and does not count them as jewelry tasks. Generic craft and workshop lines that name no jewelry task are not counted either.

The catalog describes work done with "fine tools", "pliers", "gravers" and, for engraving, "under magnification". The site does not state piece sizes or tolerances, so this page makes no claim about how small the work is. What the site does state is what egocentric video puts into the training distribution: hands entering the frame from below, occlusion during grasp and motion blur during reach.

The site also states why a hand pose stream matters here. RGB alone leaves a model to infer 3D hand structure it cannot see reliably through occlusion and motion blur. Firsthand delivers 3D hand pose as handJoints[2][21][3] in metric units in the head-camera frame, with a visibility flag per joint, aligned to depth and RGB. Two hands of 21 joints each are tracked per frame, and the flag records which of them were hidden.

The catalog places all five rows in the Jewelry & Metalsmithing Studio environment. The site names no studio, city, school or jeweler, and it does not state capture at a customer own studio or facility. Environment, access and schedule are agreed in the spec, and the number of studios is Not stated.

Every episode links to a signed contributor release through a documented consent chain. Faces, plates and screens are detected and irreversibly blurred at ingest, a reviewer confirms the result before anything ships, and the anonymization record is delivered with the data. An undocumented bystander is also a stated reject criterion, and the site states no further privacy rule specific to studios.

A head-mounted phone records the first-person view of real work on site, and a multiview rig adds synchronized third-person angles when the spec asks for them. Both are the capture modes the catalog lists for every jewelry row.

The site states a head camera at 3840 × 2160 and 60 fps for its instrumented rig, but no camera count, resolution or frame rate for the jewelry rows, so the capture tier is agreed in the spec. Throughput is Not stated, and session length and volume are quoted against your spec.

Each episode follows the Firsthand episode schema: ten streams synced under 2 ms, with head and wrist RGB, depth, 3D hand pose with per-joint visibility, contact and grasp events, and frame-accurate action segments. For bench work, hand pose tracks how the fingers hold and move the piece, contact events mark when fingers or a tool meet it, and action segments mark each step.

Object labels use the terms your spec sets. Force or tactile data, and success or quality labels for a finished piece, are Not stated. They are defined in the spec if your policy needs them.

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 any jewelry 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.

Episodes are machine-checked at ingest against the stated reject criteria, listed in the table below, and a human reviewer then checks the result. The site states them for every episode, so they apply to jewelry work too.

It does not state how often a small piece leaves the view at a bench, and it states no blur or sharpness rule. A rejected item goes to the reject log and is recaptured at our cost.

Delivery formats are RLDS, LeRobot and WebDataset, plus HDF5, zarr and Rerun .rrd. The format table below sets each against what its official documentation states. Jewelry episodes use the same schema as every other Firsthand collection, so a loader that reads another Firsthand dataset reads this one without changes.

Every batch is validated against signed accept criteria before it ships, with a reject log, and delivery is rolling, per validated batch.

Where coverage is already deep, the first validated batch lands in a median of 48 hours from signed spec. Thin environments typically take three to four weeks, because vetting operators there is the bottleneck, and we quote that up front.

On-demand top-ups run against the same spec and accept criteria after each training run. All five jewelry rows are Available on request, so the site states no capture timeline for them beyond a quote against your spec.

You describe the pieces, steps and sequence your policy needs. We turn that into a written spec covering tasks, environments, capture tier, annotation depth and accept criteria, then quote reach, timeline and price before any capture begins.

A Spec Pilot, a small and fast first batch, can prove the spec and accept criteria before you scale. Billing is per validated hour, a reject log ships with every batch, rejected items are recaptured at our cost, and every collection ships under a perpetual, buyer-owned commercial license.