Commercial collection · Region

Egocentric Data Collection in Asia-Pacific

First-person video of real commercial work across Asia-Pacific, recorded to one written spec, consented, anonymized and QA-checked before it ships.
Countries
Australia, Japan, New Zealand, Singapore, South Korea, Taiwan, India, Philippines and Vietnam
Legal basis on file
APPI · PDPA · PIPA
Settings
Commercial only
Start
Spec Pilot

Definition

What does egocentric data collection across Asia-Pacific mean?

Egocentric data is video recorded from the point of view of the person doing the work, with their hands, tools and the objects they handle in frame. Regional collection runs that capture across several countries under one spec, so a model sees the same tasks performed in different markets rather than in one.

Packaging, tools, store layouts and work habits differ between countries even inside one region. A model trained on a single country's footage quietly overfits to it. Spreading a spec across Asia-Pacific is how a program widens that variation on purpose, while the spec, the QA gates and the delivery schema stay fixed.

Countries

Which Asia-Pacific countries are in the program?

Each row links to the country page. Status and legal basis come straight from the published coverage data.

Commercial egocentric collection countries in Asia-Pacific
CountryStatusLegal basisListed environments
AustraliaListed capture countryConfirmed per engagementRecruited to spec
JapanLegal basis on fileAPPIKitchen & food prep, Retail & shelf ops, Tool use & assembly
New ZealandRecruited to specConfirmed per engagementRecruited to spec
SingaporeLegal basis on filePDPARetail & shelf ops, Warehouse & logistics
South KoreaLegal basis on filePIPARetail & shelf ops
TaiwanRecruited to specConfirmed per engagementRecruited to spec
IndiaListed capture countryConfirmed per engagementRecruited to spec
PhilippinesListed capture countryConfirmed per engagementRecruited to spec
VietnamListed capture countryConfirmed per engagementRecruited to spec

"Legal basis on file" means the consent framework is documented in the provenance section. "Listed capture country" means the country is published as a capture country without a documented basis yet. "Recruited to spec" means contributors are sourced when a spec asks for it.

Commercial settings

Which commercial environments are listed in Asia-Pacific?

Environments that are not listed for a country are still recruited to spec; listed ones are where a spec can usually start fastest.

Commercial environments by listed country in Asia-Pacific
EnvironmentWhat is recordedListed in
Retail & shelf opsShelf stocking, facing and store-floor tasks, recorded first-person with the hands in frame.Japan, Singapore, South Korea
Warehouse & logisticsPicking, packing and handling work, recorded first-person with the hands in frame.Singapore
Kitchen & food prepCommercial kitchen and food-prep work only under this program, not home kitchens.Japan
Automotive serviceService-bay tasks, recorded first-person with the hands and tools in frame.Recruited to spec in every country
Tool use & assemblyHand-tool and assembly procedures, recorded first-person with the hands and tools in frame.Japan

See the full catalog of commercial environments and tasks.

Multi-country

How does a multi-country spec work?

  1. 01 · Brief the gapTell us the tasks, commercial settings and which Asia-Pacific countries you need.
  2. 02 · Approve one specTasks, environments, capture tier, annotation depth and accept criteria apply to every country; each country is quoted for reach and timeline.
  3. 03 · Run a Spec PilotA small, fast first batch that proves the spec and QA bar hold on real footage before volume scales.
  4. 04 · Scale by countryCountries scale against the same spec, with on-demand top-ups afterward.

A median of 48 hours from signed spec to first validated batch where coverage is deep. Where coverage is thin, the timeline is quoted against the spec before capture begins.

QA

Is the QA bar the same in every country?

Gates every recording passes, in every country, before it is billed
GateWhat is checkedIf it fails
Spec fitThe take shows the task, environment and conditions the spec asked forRejected and recaptured
ConsentA signed release exists for every person in the manipulation regionRejected; never shipped
SyncStreams are present and aligned to the shared clock (rig tier)Rejected and recaptured
AnonymizationFaces, plates and screens redacted and human-reviewedHeld until redaction passes
BillingOnly validated hours are invoicedRejected hours are never billed

Yes. The gates do not change by country, which is what makes footage from several markets comparable inside one training set. Anonymization is documented on the anonymization page.

Delivery

What is delivered, and in which formats?

Footage from every Asia-Pacific country is delivered in RLDS, LeRobot, WebDataset, HDF5, zarr and Rerun .rrd, mapped to the canonical episode schema, under a buyer-owned commercial licence. See delivery formats, sample data outputs and pricing.

FAQ

Egocentric data collection in Asia-Pacific, answered.

Which Asia-Pacific countries can Firsthand collect egocentric data in?
The commercial program has country pages for Australia, Japan, New Zealand, Singapore, South Korea, Taiwan, India, Philippines and Vietnam. Recruiting runs worldwide, so a country not listed here can still be scoped against a spec.
Is there a legal basis on file for Asia-Pacific?
Yes for Japan (APPI), Singapore (PDPA) and South Korea (PIPA). For the other countries the legal basis is confirmed per engagement and written into the spec before capture starts.
Can one spec cover several countries?
Yes. Multi-country specs are normal: one written spec sets the tasks, environments, capture tier, annotation depth and accept criteria, and each country is quoted for reach and timeline against it.
What does "commercial" collection mean here?
Recordings are made in commercial settings only: retail floors, warehouses, commercial kitchens, automotive service bays and tool-use and assembly work. Household and home-routine footage is outside this program.
Is the footage anonymized?
Faces, licence plates and screens are detected and blurred in the anonymization pipeline, with a human review step, before footage is delivered.
What formats is the data delivered in?
RLDS, LeRobot, WebDataset, HDF5, zarr and Rerun .rrd, each with the same canonical episode schema, so footage from every country in a program loads the same way.
How much does collection cost?
Pricing is quoted per engagement rather than from a rate card, and billing is per validated hour: footage that passed sync and QA against your spec. See the pricing page for how quotes are built.

Sources

Where are the documented legal bases published?

Need commercial footage from across Asia-Pacific?

Tell us what your model is missing. We quote reach, timeline and price per country against one written spec, starting with a Spec Pilot.