Datasets & Deliverables

What we deliver

Research-ready dataset packs with raw footage, processed copies, manifests, and schema versioning.

Available

Raw Capture Pack

Unprocessed egocentric footage

Direct phone recordings from our contributor network. Includes original video files, device metadata, and capture timestamps. Full quality preserved.

MP4/MOV original recordings
Device orientation data
Capture session metadata
Geographic region tags

What's included

Raw video files
JSON manifests
Metadata CSV
Session logs
Pilot

Standardized Dataset

Training-ready data packs

Quality-validated footage with standardized encodes (1080p, 30fps, H.264). Includes original files plus processed versions ready for integration.

Quality-validated clips
Standardized resolution/FPS
Originals + processed included
Schema-versioned exports

What's included

Processed MP4s
Task metadata
Train/val splits
Example scripts
Coming Soon

Annotated Dataset

Temporal action labels

Verb+Noun action segments with timestamped boundaries. EPIC-KITCHENS compatible taxonomy. Currently in pilot with select partners.

Timestamped action segments
70+ verbs, 175+ nouns
EPIC-KITCHENS aligned
Human-verified quality

What's included

JSONL annotations
Action labels
Segment boundaries
Documentation

Technical Notes

Data specifications and compatibility

Capture Protocol

  • • First-person (egocentric) perspective
  • • 1080p recommended, original quality preserved
  • • 5-second scene scan before each task
  • • Landscape orientation standard

Annotation Schema

Pilot
  • • Verb-Noun action segments
  • • EPIC-KITCHENS compatible taxonomy
  • • Timestamp-based (seconds, not frames)
  • • 70+ verbs, 175+ nouns

Compatibility

  • • RGB video for VLA/behavior cloning
  • • Scene scan enables optional SfM by customer
  • • Raw files preserved for flexibility
  • • JSONL export format

Data Processing Pipeline

From raw capture to annotation-ready standardized files

01

Raw Capture

Original 4K/1080p from contributor devices

02

QC & Validation

Frame check, audio sync, content review

03

Standardization

720p H.264 30fps CFR encode

04

Delivery

Raw + Proxy + Manifests package

Standardization Specs

CodecH.264Universal playback
Frame Rate30 CFRConstant, no drift
Resolution720pAnnotation-optimized
OriginalsPreservedFull quality master

Why standardize? Raw mobile videos have variable frame rates (VFR), different codecs (HEVC/H.264), and varying resolutions. Our pipeline normalizes everything to a consistent format for reliable annotation timestamps and seamless integration with your training pipelines.

Ready to explore?

Request a sample dataset pack to evaluate our data quality and format compatibility.