Geometry Functionals#

physicsnemo.nn.functional.signed_distance_field(
mesh_vertices: Tensor,
mesh_indices: Tensor,
input_points: Tensor,
max_dist: float = 100000000.0,
use_sign_winding_number: bool = False,
) → tuple[Tensor, Tensor]#

Compute the signed distance field (SDF) for a mesh and query points.

The mesh must be a surface mesh consisting of triangles. This functional uses a Warp-backed implementation for accelerated execution.

Parameters:

  • mesh_vertices (torch.Tensor) – Coordinates of mesh vertices with shape (n_vertices, 3).

  • mesh_indices (torch.Tensor) – Triangle connectivity indexing into mesh_vertices. Expected shape is (n_faces, 3) or a flattened equivalent.

  • input_points (torch.Tensor) – Query points at which to evaluate the signed distance, with shape (..., 3).

  • max_dist (float, optional) – Maximum search distance for closest-point queries. Default is 1e8.

  • use_sign_winding_number (bool, optional) – Whether to use winding-number-based sign computation. Default is False. When False, the mesh should be watertight for reliable signs.

  • implementation (str, optional) – Explicit implementation name. Defaults to None, which uses normal dispatch (currently the Warp implementation).

Returns:

A tuple (sdf, hit_points) where: - sdf contains signed distances at each query point. - hit_points contains the closest point on the mesh for each query.

Return type:

tuple[torch.Tensor, torch.Tensor]

Examples

>>> mesh_vertices = torch.tensor(
...     [(0.0, 0.0, 0.0), (1.0, 0.0, 0.0), (0.0, 1.0, 0.0)]
... )
>>> mesh_indices = torch.tensor([(0, 1, 2)])
>>> input_points = torch.tensor([(0.5, 0.5, 0.5)])
>>> sdf, hit_points = signed_distance_field(
...     mesh_vertices, mesh_indices, input_points
... )