Data

Know exactly what the data contains.

Every licensed dataset and custom delivery is defined by its source, version, skeleton, representation, labels, quality checks, limitations, delivery structure, and permitted use.

Current studio-dataset facts

The currently licensable studio dataset contains 150+ hours of full-body human motion with finger data, recorded with a 120-camera marker-based system in a AAA production studio.

The target skeleton and delivery format can be customized for the engagement. Public examples may include UE5, MHR, and Unitree G1. Available annotations include action labels, temporal segmentation, root trajectories, joint trajectories, foot contacts, and performer/session metadata. Hand-object contact labels are not included as a standard annotation.

Commercial training and evaluation rights are available by agreement. A representative sample is available through a gated request. Those facts apply to the studio dataset; they do not generalize to video-extraction, labeling, or custom dataset programs.

What every delivery specification defines

Specification areaWhat is documented
Source and provenanceHow motion was captured or estimated, source ownership requirements, and relevant collection conditions
Dataset identityDataset name, release or delivery version, total size, inclusion criteria, and change history
SkeletonJoint hierarchy, joint names, rest pose, root convention, hand or finger coverage, and any target mappings
Coordinates and unitsHandedness, up axis, linear units, rotation representation, and transformation conventions
SamplingSource and delivered frame rate, timestamps, resampling, interpolation, and sequence boundaries
Files and packagingMotion containers, tabular or metadata files, directory structure, identifiers, and compression
Labels and metadataTaxonomy, temporal segments, trajectories, contacts, sequence fields, confidence, and allowed values where applicable
Processing historyCleanup, filtering, retargeting, normalization, coordinate conversion, and other transformations
Quality and limitationsValidation checks, exception handling, missing fields, uncertain labels, and known failure modes
Splits and versioningTrain, validation, or test split definitions; corrected-file policy; and changelog
Delivery and permitted useTransfer method and the commercial training or evaluation rights defined in the agreement

This table describes the specification categories Uthana documents. It does not claim that every field or format is present in every dataset.

Preserve source and delivery separately

The source specification records how the motion was captured or estimated. The delivery specification records what was transformed for the customer, including resampling, skeleton mapping, coordinate conversion, labeling, filtering, and packaging.

Keeping both layers explicit makes preprocessing decisions traceable, and prevents a normalized delivery from being mistaken for the untouched source data.

Marker-based capture, video-estimated motion, and customer-provided files require different specification details. Uthana does not force them into one universal format table.

Review a sample before committing

Share the target skeleton, required fields, formats, and intended training or evaluation use. Uthana will identify supported fields, required transformations, and unresolved constraints before proposing a license or pilot.