Stereo Catering and Banquet Plating Data

It is real-world, first-person video of the hand work done in a kitchen that prepares and serves food for events and institutions, recorded with two calibrated cameras so that depth can be recovered, together with 3D hand pose and the other streams the site already lists. The catalog describes the work in short phrases: portion hot dishes onto plates in sequence, wipe rims and load plates onto service trays; prep vegetables, proteins and sauces in large batches using hotel pans and commercial mixers; and set out chafing dishes, service stations and utensils, then clear and pack them after service. The buyer sets the task and the accept criteria; Firsthand handles contributors, capture, QA, consent and delivery.

The data is human capture by paid, consented contributors, not robot or teleoperation data. This page is about catering robot training data in the sense of what a policy can learn from watching a person plate, prepare and set out, and it makes no claim that any robot has been trained on it or that any robot caters anything. People reach this topic through several phrasings, such as stereo banquet plating video, an egocentric bulk food prep dataset or custom food-service data collection at scale. They describe one request from different angles.

Not stated. The catalog lists one capture setup for all three rows, a head-mounted phone or multiview rig. It does not list a stereo pair. Stereo applies here only through the site general stereo offer: the site describes a stereo head rig and a hard-environments stack, and a stereo pair for a catering order is a choice made in the skill spec and confirmed on a Spec Pilot.

This is why the page is written as a stereo and container page for these rows, not as a description of existing stereo catering footage. If you need stereo, say so in the brief. If you are happy with the catalog setup, the egocentric catering and event kitchen page describes that non-stereo route for the same three rows.

Three rows sit under Catering & Event Kitchen in the Food & Beverage sector. One is rated Easy and two are rated Medium, and every row has the status Available on request. That means the work is scoped and recruited to spec when a buyer asks for it, so this page claims no completed collection, no delivered hours and no delivered stereo footage for any of them. The first table below reproduces the three rows exactly as the catalog lists them.

The examples line for the environment reads "Hotel banquet kitchen, airline catering, hospital canteen, stadium concessions, event catering facility". These are setting types, not named venues or caterers, and the site names no caterer, venue, event or client for this environment. Other rows on the environment page are nearby context only and are not claimed here.

Banquet plating is Medium and reads "Portion hot dishes onto plates in sequence, wipe rims and load plates onto service trays." The order in the description is the useful part: portion, wipe, load, repeated plate after plate. For stereo capture that means the plate, the portioning tool, the rim and the tray all need to stay in the same view as the hands, and the step boundaries need to land on the same clock as the video.

The word hot is catalog wording. The site states no temperature, holding or food-safety rule for this row, so none is claimed here, and this page gives no recipe or serving instruction. Dish and menu types, plate counts, portion sizes and tray types are Not stated and are set in the written spec.

Bulk food preparation is Medium and reads "Prep vegetables, proteins and sauces in large batches using hotel pans and commercial mixers." The pans and mixers are tool types the catalog lists, not models. The catalog names three ingredient groups and a batch scale, not recipes, and the site does not describe a specific cut or technique in the row.

For an egocentric bulk food prep dataset, the batch scale is the point. A spec has to say whether one episode is one pan, one batch or a full prep session, what the object is when a mixer or a pan is in play, and where one step ends and the next begins. Ingredients, quantities, batch sizes and pan or mixer models are Not stated.

Event service setup and breakdown is Easy and reads "Set out chafing dishes, service stations and utensils, then clear and pack them after service." It is the only Easy row in the environment, and the one where the work is about the service area rather than the food.

Setup and breakdown are two ends of one row, so a spec can ask for either or both. Event sizes, guest counts, layouts and what is packed where are Not stated. Chafing dishes and utensils are named in the row, and the site states no model, size or count for either.

The catalog lists the three rows separately and does not state that they are performed in one event or by one worker. A spec can still name them as separate task families or chain them: prepare in bulk, plate in sequence, set out the service area, then clear and pack. That chain is a way to write a spec, not a claim about how any one event runs.

Action segments are what make a chain usable. Frame-accurate step boundaries separate portioning, wiping, loading, prepping, setting out and clearing, aligned to video and depth, so a policy or a reviewer can tell where one step ends and the next begins. How many segments an episode holds is Not stated.

This page is stereo-offer capture of the catering and banquet plating rows. The egocentric catering and event kitchen page covers the same three rows as first-person video without stereo as its subject. The egocentric meal plating and serving page is broader: it covers eleven plating, portioning and serving rows across several environments, including Banquet plating, at mixed statuses, and it is not stereo.

The stereo commercial kitchen line cooking page is stereo, but it covers two Restaurant & Commercial Kitchen rows about knife work and chopping, both Active collection, not catering. The egocentric food portioning and weighing page covers two portioning and packaging rows, one Ready to deploy and one Active collection, and it is not stereo. The second table below puts the pages side by side with their status as the site lists it.

The catalog states one mode for all three rows: a head-mounted phone or multiview rig. A head-mounted view records the first-person work, and a multiview rig adds synchronized third-person angles when the spec asks for them. The site describes two stereo configurations. The stereo head rig, dataset 06, is a dual fisheye pair at 1080p per eye on a 62 mm baseline, with per-unit calibration, a 9-axis IMU and hand-tracking output, shipped as MCAP. The hard-environments stack is a 4K/60 head camera with a 120° lens, two 1080p/60 wrist cameras, depth at 848×480 and 30 fps and a 200 Hz 6-DoF IMU. Which of these a catering order uses is Not stated and is agreed in the skill spec.

Whichever rig is chosen, calibration is measured every session and shipped with the episode, with a 0.28 px RMS reprojection gate. Hand pose is 21 joints per hand with a per-joint visibility flag, and the language data is one task instruction per episode. Camera count, resolution and frame rate for the set as quoted are Not stated.

Hand occlusion is the first case. Fingers close around a plate, a ladle or a utensil, and a hotel pan or a mixer bowl hides what the hands are doing inside it. The site states that the hands occlude the work and that fingers disappear behind objects. The per-joint visibility flag marks joints that are not visible instead of guessing them, contact events mark when a hand meets a plate, a pan, a tray or a utensil, and a spec can ask for wrist or multiview angles so a grip the head camera loses is still recorded.

Reflective surfaces are the second. The site states that depth degrades on reflective surfaces, which per-frame confidence maps handle by masking low-confidence regions. Glazed plates, steel pans and chafing dishes are the cases to plan for, and how depth behaves on a transparent lid or glass is Not stated, so a Spec Pilot checks it on real footage. Deformable items are the third: the site states that masks for deformable objects are shipped per frame rather than propagated, with instance masks stored at 848×480, and what counts as the object, the portion or the plate, is a spec decision.

Repetition is the last. Plating is the same step repeated plate after plate, which is hard to cover with a few recordings, so a spec names the number of sequences it needs. Steam, crowded prep tables and lighting are not described on the site, so they are Not stated and are fixed in the skill spec. An undocumented bystander is a stated reject criterion.

Every batch is validated against signed accept criteria before it ships, with a reject log. The site states that an episode is rejected when the manipulated object is out of frame for more than 15 percent of it, when more than 35 percent of depth pixels in the manipulation region are invalid, when a sync breach runs beyond the ceiling for more than three consecutive frames, when an undocumented bystander appears, or when the episode does not match the named skill spec. Episodes that fail are rejected and logged per batch, not delivered with a caveat.

Billing is per validated hour, and rejected items are recaptured at our cost. The site does not state an accuracy figure for step labels, and it lists no label for dish identity, portion quality, success or quality. What counts as a pass for a step boundary, such as the end of a wipe or the start of a tray load, is set in the accept criteria.

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, with the anonymization record delivered alongside. Footage that cannot be de-identified is discarded. The site states no catering-specific rule beyond this, so that is Not stated, and it is raised in the spec if it matters.

The buyer owns the dataset under a perpetual, irrevocable, non-exclusive-by-default licence, with exclusivity available, and the executed agreement controls. Physical-safety measures, PPE rules, hygiene rules and venue access terms are Not stated. The privacy and anonymization pages are linked below.

Firsthand offers MCAP delivery. MCAP is an open-source container format with its own official site at mcap.dev, a specification at mcap.dev/spec and a repository at github.com/foxglove/mcap; read those for what the container is and how it stores records, timestamps and attachments. This page describes the container only from those sources and does not state anything more about it.

The streams the site already states for a stereo capture are the stereo pair, depth with confidence maps, 21-joint hand pose, IMU, contact events, action segments, per-frame instance masks and calibration, with one task instruction per episode. Topic names, message schemas, encoding and rates are Not stated on the site and are agreed in the skill spec. Only dataset 06 is stated as shipped as MCAP; for any custom order, MCAP can be the only delivery or sit beside RLDS, LeRobot, WebDataset, HDF5 or zarr. The third table below maps each stream to what the site states.

Caterer, venue, event, client and brand names, equipment models, cover counts, service times, portion sizes, temperatures, ingredient lists, batch sizes, task times, throughput, error rates, force or tactile data and any food-safety or certification standing are not stated on the site. None is claimed here.

If your policy needs any of the fields that can be specified, they are written into the skill spec as annotation depth and covered in the quote before capture begins.

The site off-the-shelf catalog lists about 324,800 hours across 18 datasets, including about 64,800 hours of stereo video in 9 datasets. One dataset description mentions kitchens and food and coffee prep among other work, but none names banquet plating, bulk food preparation or event service setup, so no off-the-shelf set stands in for these rows. One stereo set is published as a format and rig example, and it is not coverage of any catering row.

Stereo catering collection to a spec is therefore custom collection: 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. All three rows are Available on request, so the scoped-to-spec timeline applies, and the quote states it up front.

Describe the rows you want, the steps you expect to see and where each starts and ends. Say whether an episode is one plate, one tray, one batch or a full loop of prepare, plate, set out and clear, what counts as the object, what counts as a reject, and whether you want a stereo pair, wrist views or the catalog multiview rig. 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.

Most programs begin with a Spec Pilot, a small first batch that proves the spec and accept criteria on real footage before you scale. The last table below shows what each spec element decides.