EgoView: Exclusive Manipulation Dataset
Long horizon task with hand tracking
EgoView is DeepReach’s curated egocentric manipulation dataset — long-horizon tasks segmented into clean per-action clips, with action labels and 3D hand-mesh reconstructions baked into every frame. Compared to other public egocentric sets, EgoView holds the strictest sharpness and stability bars, keeps at least one hand in the frame on 100% of clips, and centers framing for reliable model training. The reconstructions below recover hand meshes frame-by-frame and reproject them over the input so you can verify them against the original action.
Read the full breakdown →
How EgoView compares
Side-by-side measurements of EgoView against EgoVerse and EgoDex on the metrics that matter for manipulation training.
| What we measure | EgoView | EgoVerse | EgoDex |
|---|---|---|---|
| Hand visibility | Strong | Strong | Strong |
| Faces in frame (privacy) | 0% | 0% | 0% |
| Image sharpness | Strictest standard — sharpest frames | Comparable | Comparable |
| Camera stability | Tightest bound — least shake | Moderate motion | Higher pixel-level shake |
| Hand motion | Finest, most controlled | Larger, faster motions | Slower, fine motions |
| Hand position (vertical) | Centered | Lower in frame | Centered / upper |
| Hand position (horizontal) | Centered | Skewed right | Centered |
| At least one hand in frame | 100% | ~94% | ~98% |
What stands out
Where the three datasets diverge — and why it matters for training.
- Privacy is a shared baseline. All three datasets pass the no-faces check at 0% — a standard now essential across the field.
- Hand visibility is comparable across the board. Detectors show similar confidence levels on all three datasets, enabling models to rely on consistent hand signal presence.
- The differences show up in stability and framing. EgoVerse emphasizes dynamic motion with larger gestures; EgoDex focuses on fine tabletop work with camera shake; EgoView maintains tighter controls on motion, shake, framing consistency, and ensures complete hand presence.
- These aren't quality verdicts. Both alternatives offer value reflecting deliberate choices about activity capture. EgoView targets a specific spectrum for reliable model training.
Scenario Distribution
Share of qualified footage per scenario across the full vendor archive.
- Warehouses20.0%
- Manufacturing17.8%
- Retail13.7%
- Residential10.4%
- Offices / co-working8.2%
- Outdoor urban5.7%
- Educational5.7%
- Workshops4.7%
- Healthcare4.3%
- Parks / public3.8%
- Mixed indoor2.4%
- Mixed commercial1.9%
- Laboratories1.4%