Real homes and workplaces
Egocentric capture happens in real kitchens, workplaces, and streets, so bystanders, screens, and plates end up in frame. First-party capture means the provider controls how those are handled.
First-party egocentric data
First-party egocentric video data, captured by our own contributors.
First-person video is only as trustworthy as the chain behind it. Firsthand captures egocentric footage through its own contributor network, so every clip traces back to a briefed, paid contributor who signed a release before recording, and to the reviewer who accepted it.
- View
- First-person, head-mounted
- Captured by
- Our own contributor network
- Consent
- Release signed before capture
- First batch
- Median 48 h
Why it matters for egocentric data
First-person footage records more than the task.
A known operator behind every clip
The person wearing the camera is a briefed contributor in our own network who signed a release, not an anonymous uploader. Their hands and point of view are the signal you train on.
Captured for the task you named
Footage is recorded against your written spec, so the tasks, objects, and environments match your model rather than whatever happened to be filmed for another purpose.
Reviewed before it is accepted
Every recording is human-reviewed for task match, visible hands, and staged or idle footage before delivery, and you see the reject reasons as well as the accepted clips.
Definition
What makes egocentric video first-party?
Egocentric video is first-party when the provider delivering it also ran the capture: it recruited the contributors, briefed them on the task, paid them for their time, collected their signed releases, and reviewed the footage. Public first-person video and resold archives fail at least one of those tests, usually several.
For egocentric data the point is sharper than for most modalities. A head-mounted camera records the wearer’s surroundings as well as their hands, so the consent and de-identification behind each clip matter as much as the task inside it.
Capture
How is first-party egocentric video captured?
Contributors are briefed and collected against the spec through Firsthand’s distributed contributor network, which uses a web portal for assignment, upload, and review. Where the project needs it, they capture on our synchronized egocentric hardware; for routine programs, contributors wear an Insta360 on a quick-release head strap and record routines in their own homes and workplaces.
Uploads missing camera telemetry are rejected automatically, and a camera-position photo can accompany a recording to show how and where the camera was worn.
Consent
How is consent recorded for first-person footage?
Every contributor signs a release before capture granting the usage rights the buyer needs, and those consent records ship with each batch. The release text is published verbatim on the provenance page.
The release also covers bystanders: contributors confirm they have not knowingly recorded anyone who has not signed it, and where an unidentified person appears, the footage is de-identified or discarded.
Anonymization
How are bystanders removed from egocentric footage?
Faces, licence plates, and screens are detected and irreversibly blurred at ingest, then confirmed by a reviewer before anything ships. The consented operator, their hands, and their point of view are untouched, because that is the training signal.
If a person cannot be de-identified, the clip is discarded rather than shipped.
Delivery
What do you receive with first-party egocentric video?
Accepted footage ships with per-recording metadata, the consent records, and the reject log, organised in a human-readable, PII-free layout of project, category, activity, slot, and version.
Custom egocentric collections deliver in RLDS, LeRobot, and WebDataset, plus HDF5, zarr, and Rerun .rrd, with readable converter source so you can retarget the schema to your own loader. The licence is perpetual, irrevocable, worldwide, buyer-owned, and non-exclusive by default.
Getting started
How fast can first-party egocentric video be delivered?
Where coverage for your tasks and regions 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 while recruiting expands.
A Spec Pilot, a small, fast first batch against your written spec, proves the spec and accept criteria on real footage before you scale.
At a glance
What you get.
- View
- First-person, head-mounted
- Captured by
- Firsthand’s own contributor network
- Consent
- Signed release per contributor
- Review
- Human review of every recording
- Anonymization
- Faces · plates · screens, reviewer-confirmed
- Formats
- RLDS · LeRobot · WebDataset · HDF5 · zarr · Rerun .rrd
Every collection runs through the same seven-stage end-to-end custom collection pipeline. End-to-end custom data collection is a managed service that takes an AI data need from problem to owned dataset in a single accountable pipeline.
Sourcing comparison
Sources of egocentric video, compared
| Question | Public or scraped first-person video | First-party capture (Firsthand) |
|---|---|---|
| Who wore the camera? | Usually unknown | A briefed contributor in our own network |
| Did they consent to commercial training? | Check the source’s licence; often not | Signed release before capture |
| Were bystanders handled? | Varies | Blurred and reviewer-confirmed, or the clip is discarded |
| Does it match your task? | Only by coincidence | Captured to your written spec |
| Can you see what was rejected? | No | Reject log with reasons |
FAQ
First-party egocentric video data, answered.
- What is first-party egocentric video data?
- First-person footage recorded by contributors the provider itself recruits, briefs, pays, and holds a release for, captured against a buyer’s spec.
- Who records Firsthand’s egocentric footage?
- Briefed, paid contributors in Firsthand’s own distributed contributor network, who sign a release before capture.
- Are bystanders blurred in first-party egocentric video?
- Yes. Faces, licence plates, and screens are irreversibly blurred and reviewer-confirmed; clips where a person cannot be de-identified are discarded.
- Do I get the consent records?
- Yes. Consent records ship with each batch, alongside the reject log and per-recording metadata.
- Which formats does first-party egocentric video ship in?
- RLDS, LeRobot, and WebDataset, plus HDF5, zarr, and Rerun .rrd, with readable converter source.
- How fast can I get a first batch?
- Where coverage is deep, a median of 48 hours from signed spec. Thin coverage runs three to four weeks.
Request first-party egocentric video.
Tell us the tasks and environments your model needs. We will scope it against a written spec and capture it through our own contributor network.