MOVAS-4D — Synchronised Multi-View 4D Video Dataset
10,000+ hours of time-synchronised multi-camera capture that resolves single-view occlusion and recovers true 3D spatial structure across time.
Overview
Single-view capture loses the object the moment a hand closes around it. MOVAS-4D solves this structurally: 6 to 12 hardware-genlocked cameras surround the work volume, so every moment of a manipulation is observed from at least one unoccluded angle.
Calibration is re-verified per session rather than per installation, which keeps extrinsic drift below the threshold where 3D lifting starts to degrade.
Data Structure
Directory layout as delivered. Every clip is self-contained: no cross-referencing required to train on a single sample.
movas-4d/
├── sessions/{session_id}/
│ ├── cam_{00..11}/video.mp4
│ ├── calib/
│ │ ├── intrinsics.json
│ │ └── extrinsics.json
│ ├── depth/{frame}.npz
│ └── pointcloud/{frame}.ply
└── manifest.parquetAnnotation Schema
Session
Environment, rig configuration, task family.
Geometry
Per-frame calibrated camera poses and depth.
Object
Cross-view consistent instance segmentation.
Quality Gates
Published thresholds, enforced at ingest. Records that fail are rejected or flagged in the manifest — never silently included.
| Gate | Threshold | Method |
|---|---|---|
| Genlock accuracy | < 1 ms | Hardware trigger across all views. |
| Calibration RMS | < 0.5 px | Re-verified at session start. |
| Occlusion coverage | > 98% | Fraction of frames with ≥1 unoccluded view of the active object. |
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
3D/4D reconstruction
Ground truth for NeRF, 3DGS and dynamic scene reconstruction.
Occlusion-robust perception
Train models that hold object state through hand occlusion.
Ego-view supervision
Use exocentric views to supervise egocentric predictions.