MOVAS-5D — Dense Five-Dimension Annotation Layer
An annotation layer spanning object detail, camera dynamics, scene summary, visual properties and semantic context — graded by domain experts, not crowd workers.
Overview
Crowdsourced annotation fails on physical-world video for a specific reason: the labels that matter — whether a grasp is stable, whether a material will deform, whether a motion is dynamically plausible — require judgment the average annotator has no basis to make.
MOVAS routes every clip through reviewers with a relevant background, and publishes inter-rater agreement rather than asking you to take quality on faith.
Data Structure
Directory layout as delivered. Every clip is self-contained: no cross-referencing required to train on a single sample.
movas-5d/
├── annotations/{clip_uuid}.jsonl
│ # one record per dimension per segment
├── schema/
│ ├── object_material.json
│ ├── camera_dynamics.json
│ ├── scene_summary.json
│ ├── visual_properties.json
│ └── semantic_context.json
└── agreement/{batch_id}_kappa.jsonAnnotation Schema
Object & Material
Instance identity, material class, deformability, state transitions.
Camera & Motion
Ego-motion class, motion magnitude, blur and exposure events.
Scene Summary
Structured natural-language summary of what occurs in the segment.
Visual Properties
Illumination, occlusion severity, clutter density, transparency.
Semantic Context
Task intent, tool relationships, procedural dependencies.
Quality Gates
Published thresholds, enforced at ingest. Records that fail are rejected or flagged in the manifest — never silently included.
| Gate | Threshold | Method |
|---|---|---|
| Inter-rater agreement | κ > 0.82 | Computed per batch and published with delivery. |
| Double-blind rate | 15% | Independently re-annotated for agreement measurement. |
| Reviewer acceptance | < 5% | Share of applicants admitted to the reviewer pool. |
Formats & Access
Export formats
Modalities included
Cloud delivery
Direct to your S3, GCS or OSS bucket. Manifest-driven incremental sync.
Air-gapped
Physical media transfer for regulated or offline environments.
Via Movas-OS
Stream and re-annotate in place through the Movas-OS platform.
Use Cases
Curriculum construction
Filter training data by occlusion, lighting or clutter difficulty.
VLM grounding
Structured scene summaries for vision-language alignment.
Failure analysis
Attribute policy failures to specific visual conditions.