
Reconstructs a triangle Mesh from an aligned PointCloud using Open3D Screened-Poisson or Ball-Pivoting reconstruction, estimating point normals when the input lacks them and cleaning degenerate / duplicate geometry before emitting.
A single-view depth backend lifts an image into a metric point cloud, which is then surfaced into a watertight mesh for export or rendering.
A reconstructed scene cloud is re-meshed and re-projected: reconstruction yields a clean surface, and a planar projection stage consumes the result for floor / wall fitting.
Mesh with no error raised. See estimate_normals.method is a closed two-value choice and any value outside `poisson` / `ball_pivoting` aborts startup.poisson_depth) multiplies both triangle count and memory pressure.density_quantile keeps every Poisson vertex; positive values trim the lowest-density vertices below that quantile to remove noise.ball_pivoting_radii are interpreted in the cloud's own input units; an empty list auto-derives radii (1.5x and 3.0x the average nearest-neighbour spacing), so a cloud at the wrong scale silently produces a sparse or empty mesh.Mesh (zero vertices, zero faces) rather than an error.Automated release