- Ego4D's standards call for video captured with informed consent, or in public with faces and other personal information blurred.Ego4D privacy documentation
- Research datasets collected signed consent or video release forms before footage was used.Ego4D privacy and ethics statement
- Automatic blurring is used to protect bystanders in egocentric video.EgoBlur paper, arXiv 2308.13093
- Researchers propose documenting how every dataset was collected, which is why provenance matters to reviewers.Datasheets for Datasets, arXiv 1803.09010
How footage review works
Reviewers compare your recording with the project's written requirements. Each project spells out what to record, where, with what equipment and with whose permission, and your footage is checked against that list. Our hub on how recording work works covers the whole project, from the details email to payment.
Review can combine people and automated checks. Automated checks catch technical problems at scale, while people judge whether the task is complete and the permissions are in place. When a problem can be fixed, a good process asks you for a revision.
Practical checks before submission
- 01
The whole task
The recording starts before the first step and ends after the last one.
- 02
Hands and work in frame
The camera points where the work happens, so the hands and tools stay visible.
- 03
A clear picture
The task is lit well enough to see, and the footage isn't blurred by sudden head turns.
- 04
Permissions
Every identifiable person has signed a release, and a workplace has given written permission.
- 05
Nothing private
No readable screens, documents, mail or people who haven't agreed.
- 06
The project's own rules
The setting, task and handoff match what the project asked for.
For the recording habits behind each of these, read our guide to recording first-person video.
Common mistakes to check against your project
| Problem | Why it can cause problems | How to fix it |
|---|---|---|
| Task cut off at the start or end | A model can't learn steps it never sees | Start before the first step; stop after the last |
| Hands leave the frame | The work itself is missing | Tilt the camera slightly down and look at the work |
| Dark or blurry footage | Reviewers and models can't see what happens | Turn on the lights; move your head smoothly |
| A coworker or family member without a release | Their privacy wasn't protected | Get a signed release first, or keep them out of frame |
| A screen or document readable in frame | Private information leaks into the data | Clear the area or turn screens away before recording |
| Wrong task or wrong setting | It doesn't match the project | Re-read the requirements before you record |
Permission problems are the hardest to fix once you've recorded. Our guide to recording at work: privacy and consent explains releases and site permission.
How to handle a revision request
A revision request means the footage is close but something needs fixing. It comes with a deadline, so work through it like this:
- Read what the request says is wrong and fix that, leaving the rest alone.
- Re-record only what is needed, unless the request asks for the whole task again.
- Watch the new clip yourself before you send it: hands in frame, task complete, nothing private.
- Send it the way the project describes, within the time the request gives you.
If anyone asks you to pay to "unlock" a review or a payout, it's a scam. Learn how to spot AI training job scams.
How research datasets handle privacy
Large research datasets take consent in egocentric video seriously. The Ego4D privacy documentation says its "standards typically require that video be captured in controlled environments with informed consent or else in public where faces and other PII are blurred." Its consortium's privacy and ethics statement adds that "Consent forms or video release forms were signed."
EPIC-KITCHENS published its acquisition guidelines, which say "The participant should have signed the attached consent form, before recording". Its consent form says that taking part is voluntary. The EgoBlur paper describes how recorded video "is processed by an AI anonymization model" to protect bystanders.
Why quality matters to buyers
Teams that use training data need to know what they are getting. The authors of Datasheets for Datasets propose that every dataset come with a datasheet that "documents its motivation, composition, collection process, recommended uses". Incomplete tasks, missing releases or private information in frame make a dataset harder to document and riskier to use.
Buyers also judge a dataset against their own checklist, which we cover in how to evaluate video dataset quality. Gameplay projects record the screen and inputs, with no camera view, so they're reviewed against their own rules; see recording gameplay for AI.
How TaskCapture reviews footage
Each TaskCapture project lists its requirements and equipment before you accept. Footage is reviewed by people and by automated checks, and if you get a revision request, you have 5 days to send a corrected version. Review usually takes about a week, with a contractual limit of 30 days. Once your footage is accepted, you're usually paid within 7 days; the contractual limit is 60 days after acceptance. Buyers only receive footage after quality review.
Pay is USD 5–15 per hour of accepted footage. It varies by project and is not guaranteed. The Data Collector Agreement and Privacy Policy set out review and consent in full, and the collector page explains pay and payout.
Questions
What problems should I check for?
Check for missing permissions, a task cut off at the start or end, hands leaving the frame, and private information such as screens or documents in view. Your project's requirements may list more.
Can I fix footage after a revision request?
Usually, yes. The request tells you what to fix. Re-record that part, check it yourself and send it back within the time the request gives you.
Does better camera quality mean faster acceptance?
Not by itself. Reviewers check the project's requirements: a complete task, visible work, permissions and privacy. A clear, well-lit recording helps, but it doesn't replace any of those.
Why do reviewers care about releases?
Because footage of people who haven't agreed puts their privacy at risk and can't be documented as properly collected. Research datasets such as Ego4D and EPIC-KITCHENS collected signed consent for the same reason.