Skip to content

COMPANY AND PROGRAM REVIEW

ActiScale review for egocentric robotics datasets

ActiScale describes catalog data, custom collection, and continuous deliveries built from first-person human demonstrations.

By Fiund · Updated 2026-11-05

Public-source review. Fiund is a competing AI-data business. No paid engagement was tested for this review.

ActiScale describes catalog data, custom collection, and continuous deliveries built from first-person human demonstrations. Its public examples show layers such as raw video, hand information, action labels, and narration. It is worth screening when a buyer specifically needs labeled human manipulation sequences and can test what each annotation layer adds. Confirm the format and resolution on files representative of your delivery. ActiScale product and examples

Price, source and acceptance

At review, ActiScale displayed an advertised $8–$90 per-hour range across tiers and described a first packaged release. Those are vendor statements. Confirm available hours, price for the selected annotation layer, and how a deliverable hour is measured. The homepage also says collectors are paid on accepted work and invites them to a waitlist. The buyer pricing range does not give a contributor earnings rate. Ask for project eligibility, a written pay rate and rejection or appeal rules before recording. ActiScale product, tiers and collector information

Check the rights at clip level

ActiScale describes consent at capture and commercial training rights. Ask which permission covers the wearer, other people, the location, and visible protected content. Review deployment, derivative, retention and redistribution rights in the actual agreement. When pose or annotation quality matters, request a sample where those layers can be tested against the original video. ActiScale product and rights statements

An original pilot for annotation alignment

Consider a fictional 60-second tabletop sorting demonstration. The left hand briefly hides the object at seconds 18–21, and the operator corrects an initial placement at second 34. Request the original video, the corresponding annotation timestamps and a description of how hidden or uncertain frames are represented. This is a proposed review exercise, not a test performed on ActiScale data.

Check whether the hand labels follow the correct person, whether narration describes the observed correction and whether the action boundary includes the result. A smooth overlay can look convincing while attaching the wrong action to the right frame. Record disagreement separately for timing, hand identity, object contact and task outcome. Ask which layers are available in the quoted tier before treating a demo overlay as an included deliverable.

Make the hourly unit comparable

A minute of activity can have an original video, several processed versions and multiple annotation streams. List those as related files and ask how they enter the payable-hour calculation. Clarify silence, unusable views, repeated attempts, overlapping cameras and excluded segments. Keep the accepted activity duration separate from a count of derived files.

For collectors, obtain the task instructions and acceptance conditions before recording. The published buyer price does not settle what the collector receives. For labs, agree which rejection categories require replacement and retain the sample manifest used during evaluation. A successful sample review supports that sample and specification; it does not measure the complete advertised catalog.

Primary sources and further reading