Glossary
What Is Data Recollection (Re-Capture) in Custom Data Collection?
Last reviewed: October 2026
Short answer
Data recollection, or re-capture, is collecting replacement footage for episodes that failed quality checks, instead of shipping them with a caveat. Firsthand states that rejected episodes go to a reject log with reasons and are recaptured at its cost, and billing is per validated hour. A recapture turnaround is Not stated.
01
What is data recollection?
Data recollection, also called re-capture, is recording new footage to replace episodes that failed a quality check. It is not a file edit: the replacement is captured again against the same spec. That matters for robot training data because some defects cannot be repaired afterwards. The quality-gate page says timing and geometric accuracy cannot be restored in post-processing. Firsthand states that it recaptures rejected episodes at its own cost.
02
What does Firsthand state about rejected episodes?
Every episode is machine-checked, then reviewed by a person against the task spec. Episodes that fail go to a reject log with reasons. The site states that they are recaptured at Firsthand’s cost, that the reject log ships with every batch, and that each episode’s meta.json carries reject_flags. Billing is per validated hour, not per raw hour, so rejected footage is not billed. A recapture turnaround, a cap on recapture rounds, the share of episodes rejected and credit or refund terms are Not stated.
03
Why do episodes get rejected?
The capture page lists six reject reasons: a sync breach, calibration drift, occlusion, invalid depth, a consent gap and a spec miss. A sync breach is any stream above the 2.0 ms ceiling for more than three consecutive frames. Occlusion is the manipulated object out of frame for more than 15 percent of the episode, and invalid depth is more than 35 percent invalid depth pixels in the manipulation region. The first table gives each reason and what happens next.
04
How does re-capture fit into the quality gate?
A quality gate is the check, and re-capture is what a failed check leads to. The order the site describes is a machine check, human review against the spec, a reject with a logged reason, a recapture, and billing on validated hours only.
The clock for a first validated batch ends at a validated batch, not at raw footage, so the criteria in the spec decide when it stops.
05
How is re-capture different from re-annotation?
Re-capture records new footage against the spec. Re-annotation changes labels on footage that already exists. The site notes that a label error can sometimes be re-annotated, but a consent gap cannot be patched, because a release that was never signed or a face never confirmed blurred is not a quality defect to fix afterwards. Sync and calibration failures cannot be restored in post-processing either.
Whether a failure is re-run depends on whether the scene can be repeated: a sync or calibration failure on a repeatable task is typically re-captured, and a consent gap is excluded. Re-annotation as a separate Firsthand remedy is Not stated. Failing label checks are recaptured or dropped.
06
What does re-capture cost, and how long does it take?
Cost is stated: recapture is at the provider’s cost and billing follows validated hours. Time is only partly stated. The site says recapture protects delivery quality but adds time in proportion to how often it happens. It gives a median of 48 hours from a signed spec to a first validated batch where coverage is already deep, and three to four weeks where contributors must be recruited. A recapture turnaround and a reject rate are Not stated.
07
How do you request a fast, on-demand custom collection?
Request a custom dataset with your tasks, environments, delivery format and accept criteria. Collection is on demand and to a written skill spec. A Spec Pilot is the optional small first batch, and top-ups run against the same spec after each training run. Write the reject rule and the handling of unrepeatable scenes into the spec.
08
Reject reasons the site states and what happens next
| Reject reason | What the site states | What happens next | Status |
|---|---|---|---|
| Sync breach | Any stream above the 2.0 ms ceiling for more than three consecutive frames | Reject log with reason; recaptured | Stated |
| Calibration drift | RMS reprojection under 0.28 px or the session does not start; a closing solve that differs beyond tolerance is a drift reject | Reject log with reason; recaptured | Drift tolerance value Not stated |
| Occlusion | Manipulated object out of frame for more than 15 percent of the episode | Reject log with reason; recaptured | Per-frame threshold Not stated |
| Invalid depth | More than 35 percent invalid depth pixels in the manipulation region | Reject log with reason; recaptured | Stated |
| Consent gap | A bystander who cannot be de-identified | Reject log with reason; excluded, not patched | Stated |
| Spec miss | Episode does not match the named skill it was captured against | Reject log with reason; recaptured | Stated |
10
Re-capture, re-annotation and dropping an episode, as the site describes them
| Remedy | What changes | What the site states | Status |
|---|---|---|---|
| Re-capture | New footage is recorded against the spec | Rejected episodes are recaptured at the provider’s cost | Stated |
| Re-annotation | Labels change on footage that already exists | A label error can sometimes be re-annotated; a consent gap cannot be patched | Firsthand re-annotation policy Not stated |
| Dropping an episode | The episode is excluded and not delivered unflagged | Failing episodes are recaptured or dropped | Rule for choosing Not stated |
| Rejecting a delivered batch | Buyer refuses a delivery | No sign-off step, review window or dispute process described | Not stated |
11
Frequently asked questions
- What is data recollection?
- Recording replacement footage for episodes that failed a quality check, instead of shipping them with a caveat. It is also called re-capture or recapture.
- Does Firsthand recapture rejected episodes?
- The site states that rejected episodes go to a reject log with reasons and are recaptured at the provider’s cost.
- Do I pay for recaptured episodes?
- Billing is per validated hour, not per raw hour, so rejected or off-spec footage is not billed. Credit or refund terms are Not stated.
- How long does a recapture take?
- Not stated. The site gives a median of 48 hours to a first validated batch where coverage is deep, and three to four weeks where contributors must be recruited, but no recapture turnaround.
- What share of episodes are rejected?
- No general reject rate or pass rate is stated on the site, so this page does not supply one. Put the rate you will accept in the spec.
- How do I ask for recapture terms in a custom collection?
- Request a custom dataset and write the reject rule, the replacement expectation and the handling of unrepeatable scenes into the spec before capture.