Egocentric Watch and Small Mechanism Repair Data

Egocentric watch and small mechanism repair data is real-world, first-person video of a person repairing a watch: opening the case, replacing the battery or movement parts under magnification and resealing it. It is recorded from the operator viewpoint, with depth, 3D hand pose, contact events and frame-accurate action segments aligned on one hardware clock.

Repair differs from assembly in its order of work. A repairer starts with a closed object, takes it apart in a set order, swaps one part and puts it back together, so the order of the steps and the hands that carry them out are the content. The device repair page covers phones, laptops and consumer electronics. This page covers the one repair row the site states for watches.

The site states one watch row and no other watch, clock or small-mechanism row. This page says so plainly instead of inventing a task list. Anything beyond that row is a spec conversation, not a catalog entry.

The task catalog states one watch task. It is called Watch repair, it sits in the Repair Services sector and it is listed in a Repair Shop environment that covers appliance, bicycle, shoe, watch and furniture work. The catalog describes it as "Open the case, replace the battery or movement parts under magnification and reseal." It is rated Hard and listed as Available on request.

The same environment holds five other rows: Appliance repair, Bicycle repair, Electronics repair, Furniture restoration and Shoe repair. They are neighbours, not watch tasks, and the second table below lists each with the reason it is not counted. A separate Electronics & Device Repair Shop environment lists a single Device Repair row, which is also a neighbour.

The row names three steps in one sentence. First, open the case. Second, replace the battery or movement parts under magnification. Third, reseal. The first table below sets each step against what the site does and does not state about it.

The row says "the battery or movement parts", so a spec can choose battery replacement, movement-part replacement or both. The site does not state which one a collection covers, how many parts are replaced in an episode, or what counts as a finished repair.

The phrase "under magnification" appears in the row as part of the task. The site does not state a magnified camera view, a macro lens or any close-range capture setting for repair, so this page does not claim one.

No. The site states no clock row, no clock-movement row and no row for any other small mechanism. The words small mechanism in this page title describe the kind of task a spec may name. The only stated catalog row is the watch row, and the page is built on that row alone.

If a policy needs a different small mechanism, describe it in the spec: the object, the steps, and where the sequence starts and ends. The site does not state which small mechanisms it has collected, so none are listed here.

The Watch repair row sits in the Repair Shop environment, whose examples are electronics repair, appliance workshop, bicycle shop, shoe repair, watch repair and furniture restoration. The environment is listed under the Repair Services sector.

The Watch repair row is Available on request, which the site defines as scoped and recruited to spec when a buyer asks for it.

The row uses the head-mounted phone capture mode, which the site describes as the first-person view of real work on site. So the site describes this as work captured in a working environment, not a staged set, and the row does not use the staged handheld-gripper mode. The site does not state which sites a collection runs at, does not name a shop or a watchmaker, and does not state capture at a buyer own bench or shop, so the environment, access and schedule are agreed in the written spec.

The Watch repair row names the capture mode Phone (head-mounted) / multiview rig. 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. A wrist-mounted camera is available alongside the head camera for close-range detail.

Depth, 3D hand pose, contact events and action segments are aligned to video on one hardware clock, synced under 2 ms. Distance-invalid pixels beyond the depth sensor reliable range are flagged rather than interpolated. The site does not state how depth behaves at the close range of a bench, so raise that in the spec if your policy relies on it.

The site does not state a camera count, resolution or frame rate for repair. Lighting, bench tools, watch types and part sizes are Not stated, so they are agreed in the written spec.

Each episode follows the same episode schema used across Firsthand custom collections: 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 watch repair, hand pose tracks the operator through each step, contact events mark when the fingers make and break contact with the case, a tool or a part, and action segments mark the start and end of each step. Because the labels share one clock with the video and depth, a step boundary lines up with the exact frame where it happens.

The site does not say how finely hand pose resolves very small parts, so this page makes no precision claim.

A repair is a sequence of small manipulations in a fixed order, and a policy has to learn both the moves and the order. First-person data gives the hand and object detail for each move, and the action segments give the order. That combination is what the site describes for every task family it collects.

The dexterous manipulation page covers per-finger hand pose and contact events, which is the hand-level data a dexterous hand needs. This page applies the same streams to one repair sequence.

The site states no robot results for watch repair and makes no claim that any model was trained on this data. This page describes what is collected, not what a policy trained on it will do.

Delivery formats are RLDS, LeRobot and WebDataset, plus HDF5, zarr and Rerun .rrd. The first three are the training-loader formats; HDF5 and zarr carry arrays for custom pipelines, and Rerun .rrd gives a multi-stream synchronized playback file for review. The last table below sets each against what its official documentation states.

Converter source is readable, so you can retarget the schema to your own loader, and the episode schema page lists the fields.

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

Every episode links to a signed contributor release through a documented consent chain, and the consent record is part of the delivery. Contributors are the people who perform the repair, so their release covers the footage captured of their hands and workspace.

Faces, plates and screens are detected and irreversibly blurred at ingest, and a reviewer confirms the result before anything ships. The anonymization record is delivered with the data.

Text on a watch case, a dial or paperwork, and engraved or serial markings, are not listed as blur categories on the site, so this page does not claim they are blurred.

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 timeline up front.

The Watch repair row is listed as Available on request, which means contributors are scoped and recruited to your spec when you ask. The site does not map that status to a specific number of days or weeks, so the timeline is quoted against your spec. It does not state a number of watch contributors or a repair-specific turnaround either.

On-demand top-ups run against the same spec and accept criteria after each training run, so new repair data stays consistent with what you already hold.

You describe the repair sequence your policy needs: the object, the steps, the start state and the end state. We turn that into a written spec covering tasks, environments, capture tier, annotation depth and accept criteria, then quote reach, timeline and price against that document before any capture begins.

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