Datuva by Zovda · Training data

Video data, annotated from clip to action.

Short-form and brand video, vlog, 4K stock and camera-movement footage, sports clips with three-level temporal annotation, and egocentric sessions with seven annotation types — each item shipped with structured metadata.

Request the catalog See all corpus families

Seven annotation layers
Three-level temporal labels
Reply within 1 business day

Collection contents

What ships in this collection.

Short-form & brand

Metadata on every item

Short clips and brand films, each with duration, resolution, content type, and collection ID.

4K & camera movement

Footage sorted by shot type

4K stock footage plus FPV, aerial, and handheld camera-movement sets, organized by shot type.

Sports

Three temporal levels on one timeline

Clip, phase, and action annotated on one timeline, with timestamp tables in JSONL.

Egocentric sessions

Seven annotation types per session

Master video, proxy, and annotated preview per session: actions, arm pose, camera motion, hands, objects, hand–object interaction, and keysteps.

Video SFT

Clip-level instruction pairs

Clip-level instruction pairs for video-understanding training.

Record structure

What each record holds.

Annotation layer	What it marks
`clip · phase · action`	Three nested temporal levels, each with start and end times.
`arm_pose`	Arm posture spans within a session.
`camera_motion`	Movement type per span.
`hands · objects`	Hand and object references tied to the same timeline.
`hand_object_interaction`	Contact and manipulation spans.
`keysteps`	Procedure steps, in order, inside a session.

Training stage

Recommended use by stage.

Stage	How it is used
Continued pre-training & alignment	Footage and egocentric sessions with synchronized annotations.
SFT	Clip-level instruction pairs for video understanding.
Evaluation	Held-out clips with reference labels.

Delivery

One structure per collection.

Layer	Contents
manifest.json	Every file with name, size, date, and collection ID.
UTF-8 JSONL	One record per line for text and annotations.
Documents	Data card, field schema, and a PDF spec for each collection.
Original + proxy media	Master files at capture resolution, plus proxy files for review.

Request access

Request the video catalog.

Tell us the source type, annotation layers, and volume. We reply within one business day. Sample packs, data cards, and field schemas are shared after your request is reviewed.

FAQ

Before you request a collection.

Are timestamps frame-accurate?

Annotations carry start and end times in seconds. Granularity is confirmed per collection before delivery.

Can we take one source type?

Yes. Each source type is a separate collection, so projects often start with one and extend after acceptance.

Can the annotation schema change?

Annotation fields, sampling, and delivery format can be adjusted to your pipeline.