- UniScale // 2602.23224
- GPU
- WAVE 6
URD
THE RECONSTRUCTOR
URD produces metric-scale 3D reconstruction from multi-view images in a single feed-forward pass. It jointly estimates camera intrinsics, extrinsics, metric depth, and dense point maps — no COLMAP, no Structure-from-Motion preprocessing, no iterative optimization. Traditional 3D reconstruction pipelines are slow and fragile — URD replaces multi-stage processing with a single network call, enabling instant 3D scene understanding. Designed for resource-constrained robotic platforms that need rapid environment assessment and real-time 3D mapping.
MODULE STATUS: DEVELOPMENTMethod
1PASS
- DIVISION
- ANIMA
- WAVE
- W6
- DOMAIN
- SLAM & 3D
- WAVE 6 // ANIMA SUITE
- FOUNDATION — SCALE-AWARE 3D RECONSTRUCTION
3D RECONSTRUCTION IS TOO SLOW
Traditional 3D reconstruction needs COLMAP, SfM, and iterative optimization — minutes to hours per scene.
Robots need instant 3D understanding for real-time decision making.
WHAT URD DELIVERS
URD produces metric-scale 3D reconstruction from multi-view images in a single feed-forward pass, jointly estimating cameras and dense point maps.
CAPABILITIES
- Single feed-forward pass 3D reconstruction
- Joint camera + depth + point map estimation
- No COLMAP, no SfM, no iteration
- Metric-scale output for robotics
WHY THIS IS HARD
Building URD requires solving multiple coupled problems:
- 01Joint estimation of intrinsics, extrinsics, and depth in one pass
- 02Metric scale without known cameras
- 03Dense point cloud from limited views
- 04Real-time reconstruction speed
URD solves these through careful architecture design and rigorous validation.
PROOF, NOT PROMISES
Key metrics:
| METRIC | VALUE |
|---|---|
| Method | Feed-forward |
| COLMAP | Not needed |
| Scale | Metric |
| Output | Dense points |
WHAT'S BUILT TODAY
| COMPONENT | STATUS | NOTES |
|---|---|---|
| Feed-Forward Arch | COMPLETE | Single-pass network |
| Joint Estimation | COMPLETE | Camera+depth+points |
| Point Generation | IN PROGRESS | Dense point maps |
| Edge Deploy | PLANNED | MLX inference path |
| Core models | COMPLETE | 3D reconstruction validated |
| API layer | IN PROGRESS | Reconstruction service |
WHERE URD DEPLOYS
- APP_01
RAPID ASSESSMENT
Instant 3D scene understanding for emergency response.
- APP_02
DIGITAL TWINS
Quick 3D models from a few photos.
- APP_03
ASSET CREATION
3D reconstruction for simulation environments.
UNDER THE HOOD
FOUNDATION: UNISCALE
- Single feed-forward pass 3D reconstruction
- Joint camera + depth + point map estimation
- No COLMAP, no SfM, no iteration
KEY INNOVATION
URD produces metric-scale 3D reconstruction from multi-view images in a single feed-forward pass, jointly estimating cameras and dense point maps.
DEPLOYMENT
- REST API
- Docker containerized
- Prometheus metrics
- Configurable backends
COMPUTE
- PRIMARY
- GPU
- EDGE
- Optimized inference
- API
- REST + streaming
PAPERS
- [01]UniScale (2602.23224)