geometry.to_sdf

Read an existing closed mesh back into a distance field, so a modelled or imported shape can be smooth-unioned, shelled, repeated or displaced like any sdf.* primitive. The mesh is sampled onto a Grid (grid.box) on the device: every cell centre takes its distance to the nearest triangle, signed by the winding number (negative inside), and the grid is read back the way sdf.from_grid reads one. The grid's size is known before anything is allocated, and the work is bounded: a request past max_samples or max_work is refused naming the resolution that fits. A geometry with no triangles, or with bounds of no extent, is refused.

Inputs

• geometry (Surface). The closed mesh being sampled.
• resolution (Float, default 32, at least 2). How many cells span the longest side of the sampled box. The field carries no detail finer than one cell, so a shape with thin walls needs enough cells to hold them.
• padding (Float, default 0.1). How far the sampled box extends beyond the mesh's own bounds (at least a thousandth of the longest side). Outside the sampled box the field reports the distance to the box plus the value at its wall.
• max_samples (Float, default 300000). How many grid cells may be sampled: the memory bound.
• max_work (Float, default 1000000000). How many cell-against-triangle tests the sampling may perform: the time bound.

Outputs

• sdf (Sdf). The distance field, as sdf.from_grid answers it.
• grid (Grid). The sampled grid, its points carrying distance.
• info (Data). samples, divisions ({ x, y, z } cells), cellSize, triangles, distanceTests and bounds ({ min, max } of the sampled box).

local field = g.geometry.to_sdf({ geometry = g.mesh.box({}), resolution = 16 }).sdf