vid2scene
Phone video → metric 3D reconstruction → a room a robot can explore.
A benchmarked scene (Replica room0): original video (left) and our reconstruction (right), along the same path. Sub-centimetre accuracy against the ground-truth mesh.
Interactive viewers (live)
4 methods + consensus
PGSR · DN-Splatter · MonoSDF · the consensus fusion, aligned in one frame. Toggle to compare.
Open viewer →Reconstruction vs ground truth
Our reconstruction toggled against the Replica ground-truth mesh (Chamfer 0.62 cm).
Open viewer →Quality vs ground truth (Replica, visibility-culled · cm / F-score@5cm)
| Method | Accuracy ↓ | Completion ↓ | Chamfer-L1 ↓ | F-score ↑ |
|---|---|---|---|---|
| PGSR | 1.13 | 7.60 | 4.37 | 0.898 |
| DN-Splatter | 0.57 | 6.14 | 3.36 | 0.936 |
| Consensus (ours) | 1.07 | 6.47 | 3.77 | 0.913 |
5-scene average. Sub-centimetre accuracy. Full protocol + per-scene table in the repo's docs/BENCHMARK.md.
Interior fly-throughs of three reconstructed benchmark scenes (living room, bedroom, study).
An agent explores it
The metric mesh is loaded as a Habitat navmesh. Top-down view of a recorded agent trajectory over the reconstructed room. The Genesis physics upgrade is described in docs/PHASE2_GENESIS.md.
Code
Repo & full write-up on GitHub — see the repo, docs/ARCHITECTURE.md, and docs/BENCHMARK.md.