Egocentric Knifemaking and Bladesmithing Data

Egocentric knifemaking and bladesmithing data is real-world, first-person video of a person making blades in a metal workshop. The catalog describes the task in one line: forging, grinding, heat treating and finishing blades in a metal workshop. The video is recorded from the worker viewpoint, with depth, 3D hand pose, contact events and frame-accurate action segments aligned on one hardware clock, against a spec the buyer writes.

This page treats blade making as a metal craft and a shop task, the way the catalog does. It covers one listed row and says plainly where the site stops. It does not describe tooling, steels, temperatures or outcomes, because the site states none of them. If you are training a policy on shop skills that depend on a long chain of hand-tool steps, this page sets out what is on offer, what is quoted against your spec and what you will need to define yourself.

The sibling welding and metal fabrication page covers the other workshop rows. This page exists because the knifemaking row is rated Hard, is not yet set up for immediate capture and is the one row in the workshop that names forging and heat treating at all.

The task catalog lists exactly one row for this work, named "Knifemaking / Bladesmithing". It sits in the Metal Fabrication & Welding Workshop environment, whose example setting is a metal shop with welding bays, grinders, cutting and fitting benches. The row description reads "Forging, grinding, heat treating and finishing blades in a metal workshop." Its skill rating is Hard, its capture mode is a head-mounted phone or a multiview rig, and its status is Available on request.

Available on request means the contributors and site access for the row are not yet set up for immediate capture, so the timeline is quoted against your spec instead of being fixed in advance. The other ten workshop rows are listed as Ready to deploy, which means capture starts once a spec is signed.

The catalog does not split the row into separate forging, grinding, heat treating and finishing tasks. This page names the four stages because the row description names them. It does not claim that each stage is a separate listed task, or that every episode contains every stage. Which stages are in scope for a dataset is decided in the spec.

Four of the eleven workshop rows are rated Hard and the other seven Medium, and none is Easy. Knifemaking is one of the four, alongside the two welding rows and the metal fabrication row. The rating is the catalog skill rating for a person doing the task; the site does not translate it into any statement about how hard the task is for a policy to learn.

The row description lists forging, grinding, heat treating and finishing as one run of work on a single blade. A first-person recording suits that shape: it keeps both hands, the workpiece and the tool in one view, so the dependency between steps stays visible in a single episode. Frame-accurate action segments mark the boundaries between steps, so a training pipeline can cut an episode at the point a stage changes.

A per-joint visibility flag on the 3D hand pose marks which joints were visible in a frame, which matters when a hand wraps a handle or a piece of stock. The site does not state how many episodes a blade needs, how long a stage takes, or what a successful result looks like, so none of those appear here.

Two grinding rows sit in the same workshop: "Grinding & Deburring", described as "Grind welds and cut edges smooth and remove burrs with grinder and hand files", and "Grind and deburr welded seams using an angle grinder and files". Both are rated Medium and listed as Ready to deploy. Their descriptions are about welds, seams and cut edges, and the site does not say they are blade tasks. They are separate listed rows and this page does not treat them as knifemaking sub-tasks.

The Jewelry & Metalsmithing Studio environment lists a "Polishing" row, described as "Polish metal pieces on the wheel with compounds and finish by hand", rated Medium and Available on request. It concerns metal pieces in a studio, not blades. The sanding and surface finishing page covers polishing, buffing and deburring across environments, and the jewelry page covers the studio.

The practical consequence is that the catalog has no row for blade grinding or edge finishing on its own. Those words appear only inside the one knifemaking description. If your policy needs a grinding or finishing stage, name it in the spec and it is quoted as part of the knifemaking row, or as one of the neighbouring rows if that fits your task better.

The catalog places the row in its Metal Fabrication & Welding Workshop environment. A head-mounted phone records the first-person view of real work on site, and a multiview rig is the alternative capture mode the row lists. A wrist-mounted camera is available alongside the head camera for close-range detail of the hands and the tool.

The site does not state a camera count, resolution or frame rate for this row. It does not say how a recording handles bright light, reflective metal or a dark corner of a shop, how depth behaves on polished steel, or how many workers or sites are available. These are Not stated, so raise them in the spec before capture.

The site also does not state capture at a customer own facility for this row. Environment, access and schedule are agreed in the spec. The site names no shop, brand or worker.

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. Each episode also carries one task instruction, which is the only language data the site states. For blade work, hand pose tracks how stock and tools are held, contact events mark when fingers meet a tool or a workpiece, and action segments mark each stage.

Object labels use the terms your spec sets. Tool and machine models, steel or material types, process parameters, number of sites or workers, throughput, force or tactile data and a task-completion label are Not stated. They are defined in the spec if your policy needs them. No claim is made about temperatures, hardness, sharpness, edge retention, safety, protective equipment or compliance.

Standard Firsthand capture is cameras, depth and IMU. A separate glove tier records a tactile signal, and the tactile glove page describes it. The site does not state that the glove tier is available for the knifemaking row, and it states no pressure, force or grip value for grinding or any other stage.

If grip pressure on a handle or a tool matters to your policy, say so in the spec so the glove tier can be scoped and quoted. Do not assume standard capture carries a force signal: it carries video, depth and motion from the IMU, plus the hand pose and contact events derived on the same clock.

The welding and metal fabrication page covers the rest of the workshop: layout, pipe cutting and fitting, welding, grinding and deburring, and weld inspection. It mentions the knifemaking row as one line among eleven. This page takes that one row as its subject and sets out what the site states about it.

The sanding and surface finishing page covers repeated finishing passes such as sanding, buffing, polishing and deburring across several environments. The jewelry and metalsmithing page covers a studio bench with small pieces. Those pages describe different rows in different settings. If your task is a long chain of shop steps on one workpiece, this page is the closest match. If your task is a single finishing pass, one of those pages is likely a better fit.

Name the stages you need, using the four the row description gives: forging, grinding, heat treating and finishing. Say whether each episode covers one stage or a run of stages, and say where an episode starts and ends, so that action segments can be cut to match. Name the neighbouring rows you also want, such as the grinding rows, so they are quoted as separate listed rows.

Say which tools and machines your policy must handle, since the site states none of them for this row. Say whether you need the wrist camera for close-range detail, whether you need the glove tier for a tactile signal, and what accept criteria a batch must meet before it ships. A reject log and recapture at our cost apply to every batch, so the accept criteria are what make a rejection precise.

A short spec is fine for a Spec Pilot. The pilot exists to find out what the spec leaves open, before you pay for volume.

The page makes no claim about forging temperatures, heat treatment, hardness, sharpness or edge retention. It names no steel grade, tool brand, machine model or customer. It gives no cycle time, quality rate, defect rate, success rate or worker count, and it does not say a policy trained on this data will perform any task autonomously.

It makes no safety, protective equipment, standards or certification claim. It treats blade making as a metal craft and shop task, and it says nothing about use of the finished piece. Where the site is silent, the page writes Not stated and leaves the question to the spec, so a reader can see which parts are stated and which are open.

Every episode links to a signed contributor release through a documented consent chain, and the consent record is part of the delivery. 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.

The site does not state a separate rule for other text in a shop, such as drawings, labels, maker marks or machine displays, so this page does not claim they are blurred. It also states nothing about the ownership of a design or a finished piece. The license covers the data. Raise anything else in the spec if it matters.

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. Knifemaking 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. Because this row is Available on request, its timeline is quoted against your spec rather than assumed.

On-demand top-ups run against the same spec and accept criteria after each training run. The site does not state a batch size or later-batch turnaround, so those are quoted against your spec.

You describe the stages, tools 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.