Stereo Carpentry and Framing Data

It is first-person video of a person doing carpentry or framing work, recorded from their own point of view with two calibrated cameras and with hand pose and sensor streams alongside. The tasks are the ones the site’s catalog lists for carpentry, framing, and measuring and cutting lumber. Stereo describes the capture, and MCAP describes the container. They are separate choices.

The site does not state that stereo is already running on these rows. The catalog lists a head-mounted phone or a multiview rig as the capture setup, and a stereo pair is a capture choice agreed in the skill spec and confirmed on a Spec Pilot. The data is human capture by paid, consented contributors, not robot or teleoperation data. Firsthand is not affiliated with Foxglove or the MCAP maintainers.

The egocentric woodworking and carpentry page covers the tasks: all eleven workshop rows, the benches, the episode contents, the labels, privacy and delivery. The furniture assembly page covers gluing, clamping, drilling and joining furniture frames. This page does not repeat either. It takes the cutting and framing rows from two environments and covers the two choices layered on them: a calibrated stereo pair for depth, and MCAP as the delivery container.

The stereo construction crews page covers the Active Construction Site environment as a whole, where carpentry is one trade among fifteen task rows. Here the carpentry and cutting rows sit together, because a construction-robotics or trade lab wants site framing and workshop cutting in one specification. Sanding, staining, veneer and hardware rows stay on the woodworking page. No lumber grade, board size, blade specification or building code appears on this page, because the site states none.

Five rows count, in two environments. Active Construction Site lists Carpentry, which measures, cuts and nails or screws framing, sheathing and trim on site, Construction / Woodworking, which frames, sheathes and fits timber with hand and power tools, and a lumber row that measures and marks with a tape and cuts to length with a circular saw. Carpentry & Furniture Workshop lists Wood Cutting and a panel row that cuts on the table saw and shapes curves with a jigsaw. The first table below reproduces each as the catalog states it.

All five are marked Active collection and use the same listed capture setup. Four are rated Hard and the lumber row is rated Medium. The site names no building, brand, carpenter, lumber product or tool model for any row, so this page names none.

Carpentry is about where things are relative to each other: where a line is on a board, where a blade meets that line, where a cut piece sits against the piece beside it. A stereo pair recovers depth from the calibrated frame pair rather than from a standalone depth sensor, so the shape of the work sits in the same frame as the hands. The site gives no depth accuracy figure for these rows and no cut-quality label, so what a buyer does with that depth is their pipeline’s question.

The site’s stereo material also says that in a glare-heavy or low-texture scene a single-frame depth guess gets worse, and that a calibrated pair measures disparity between two views at a fixed, known baseline instead. It adds that stereo keeps working outdoors where active depth sensors struggle. Whether a framing job is indoors or outdoors is not stated beyond the environment names.

The rows describe measuring and marking lumber, cutting it to length, measuring and marking panels, shaping curves and fitting timber. Depth from a stereo pair puts the board, the panel and the surface under them in three dimensions, so how a piece is supported, how it is held and how far the tool is from it are in the capture. Whether a cut line is a marked line, an edge or a guide is not stated.

The site lists no label for the wood or the cut itself: no species, no board or panel dimension, no cut-line deviation or tolerance, no joint-fit measure. It states that metric depth on sub-centimetre parts is flagged rather than interpolated, and does not say whether a saw kerf, a nail or a screw head counts as such a part. If a policy depends on any of those, say so at the start so the capture is planned around it.

The site states 3D hand pose of 21 joints per hand with a per-joint visibility flag, and contact and grasp events that mark when fingers meet a tool or a part. In carpentry that is the hook and pull of a tape, the grip on a circular saw, the hands guiding a panel through a table saw, the fingers steadying a board while the other hand works and the hold on a jigsaw while it follows a curve.

These are the moments a head camera loses most often, because the tool or the board sits between the lens and the fingers. A spec can ask for a wrist view so a grip the head camera loses is still recorded, and the visibility flag marks the joints that were hidden. The site states no force, torque or tool-pose stream, and does not say whether a board, a saw or a marked line counts as the object in a contact event, so settle that in the spec.

The rows describe a few clear sequences. Measure and mark lumber with a tape, then cut it to length with a circular saw. Measure and mark panels, cut them on the table saw, then shape curves with a jigsaw. Measure, cut, and nail or screw framing, sheathing and trim. Frame, sheathe and fit timber. Each verb can be its own labeled span, which is what action segments are for.

A decision comes before capture: what counts as one attempt. It could be one cut, one panel, one framed piece or one measure-to-fit sequence. The site does not set that length, so the skill spec does, along with which verbs and nouns the segments use. The site also states one task instruction per episode, which is the only language data it describes for an episode.

The site describes two stereo setups and does not say which one a carpentry order uses. The hard-environments stack is a global-shutter 4K/60 head camera with a 120° lens, two 1080p/60 wrist cameras, a depth sensor at 848×480 and 30 fps, and 200 Hz 6-DoF IMUs. The off-the-shelf stereo head rig, dataset 06, is dual fisheye cameras at 1080p per eye, a 62 mm baseline, per-unit calibration, a 9-axis IMU and hand-tracking output.

The catalog lists a head-mounted phone or multiview rig for all five rows. Which rig, resolution and streams a carpentry order uses is Not stated, and the second table below sets what is stated beside what is not. Intrinsics and extrinsics are re-solved at the start and end of every session, and streams are hardware-triggered onto one clock with a 2.0 ms cross-stream ceiling. Capture at your own facility is listed as not offered.

Glare and low texture are the first risk. The site notes that glare-heavy and low-texture scenes make single-frame depth worse, and that depth degrades on wet and reflective surfaces while confidence maps mask the low-confidence regions. An episode is rejected when more than 35 percent of depth pixels in the manipulation region are invalid. The site says nothing about sawdust in the air or on a surface, and states a low-light share only for its automotive dataset, so dust and lighting handling for carpentry rows are Not stated.

Occlusion and motion blur are the second. The head camera is global-shutter, which avoids rolling-shutter skew, and the site notes that reach and grasp frames are where blur and occlusion show up. Hand pose carries a visibility flag for joints a tool or a board hides. Calibration is the third, because stereo depth is only as good as the rectification behind it: it is measured at the start and end of every session under a 0.28 px RMS gate.

Framing and bystanders are the fourth. An episode is rejected when the manipulated object is out of frame for more than 15 percent of it, or an undocumented bystander appears. The third table below sets each risk beside the site’s handling. The site does not describe how a fast-moving blade or a board held in a clamp is handled.

Every batch is validated against signed accept criteria before it ships, a reject log ships with it, and rejected items are recaptured at our cost. Delivery is rolling, per validated batch, and billing is per validated hour. Each 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. Footage with bystanders who cannot be de-identified is discarded, which matters on a site where other trades work nearby. The site makes no safety, protective equipment, guarding, licensing, certification or compliance claim for any carpentry or construction row, and neither does this page.

MCAP is an open source container file format for multimodal log data, supporting multiple channels of timestamped pre-serialized data. It is serialization-agnostic, so messages can be Protobuf, DDS (CDR), ROS, JSON and more. Its design is row-oriented and append-only, it stores schemas alongside the data, and it offers an optional index plus optional LZ4 or Zstandard chunk compression, with optional chunking and CRC checksums.

Native reader and writer libraries exist for C++, Go, Python, Rust, Swift and TypeScript, and the mcap CLI can inspect, merge and split files. The foxglove/mcap repository is MIT-licensed; that is the MCAP project’s licence, not the licence of a Firsthand delivery. The MCAP hub page and the glossary entry cover the format in full.

Firsthand offers MCAP delivery, and dataset 06 is stated as shipping as MCAP. For carpentry the site states a stereo pair, a stereo-derived depth stream with confidence maps, hand pose, IMU, contact events, action segments and calibration elsewhere. What it does not state is which of those are channels in an MCAP file, how they are named, what encoding they use or at what rate.

No topic name, message schema or rate is stated on this page, and none should be assumed. Name the streams that matter in the skill spec and confirm them on a Spec Pilot batch. The fourth table below sets each stream beside what the site states about it.

The site states delivery in RLDS, LeRobot and WebDataset, plus HDF5, zarr and Rerun .rrd, and MCAP can ship alone or beside any of them. Carpentry episodes follow the same episode schema as every other Firsthand collection, so a loader that reads another Firsthand dataset reads this one without changes. In the HDF5 layout the site documents, hand pose sits at annotations/hand_joints and action segments at annotations/action_segments.

Which container suits a lab depends on its reader and its training loop, and the site does not rank them. The compare page on MCAP and HDF5 sets the two side by side, and the delivery formats guide covers the rest. Say in the spec whether MCAP is the only delivery or sits beside a training format.

Off-the-shelf covers what is already in stock: the stereo category lists the stereo datasets and hand-tracking video, and only dataset 06 is stated as MCAP. This page does not claim off-the-shelf coverage of the carpentry rows. A lab can use the stock stereo footage to check the rig and the format first, then commission the carpentry tasks it needs. Custom collection covers the rest: custom collection at scale through a network of thousands of capture providers, for a consistent, recurring data flow to labs.

Where coverage is deep, the median time from signed spec to first validated batch is a median of 48 hours; where it is thin, turnaround runs three to four weeks. The construction guide describes construction and skilled trades as a thinner-coverage setting, so plan for the longer figure on site rows and ask which applies to the workshop rows when you request the quote. Collections are billed per validated hour. The supply page covers how a recurring flow is set up.

Describe the sequence your policy needs: for a cut, measure and mark with a tape, then cut with a circular saw; for a panel, mark, cut on the table saw and shape with a jigsaw; for framing, measure, cut and nail or screw. State what counts as a completed step, the setting, and the variation your pipeline is missing. Then name the streams you want in the file, the topic naming and message encoding your reader expects, whether calibration ships inside the file or beside it, and whether MCAP is the only delivery or sits beside RLDS, LeRobot or WebDataset.

Most programs begin with a Spec Pilot, a small first batch that proves the spec and accept criteria on real footage before you scale. Lumber grades, dimensions, blade specifications and tool models are agreed in the spec because the site does not state them. The fifth table below shows what each spec element decides.