nx_visual_hull.nx

nx_visual_hull.nx -- SILHOUETTE -> 3D FORM. Builds a signed field by intersecting the back- projections of N silhouettes (the visual hull), which surface_nets then polygonizes into a mesh. This is the missing rung between a photograph and geometry. The ecosystem could decode JPEG (nx_jpeg_decode_rgb), polygonize a field (nx_isosurf), render a mesh (nx_trimesh) and import OBJ (nx_obj_import) -- but had no way to turn images INTO a field. Verified absent before writing: every `carve` in the corpus is terrain/geology. ★VIEWS MUST DIFFER IN ANGLE. An anterior and a posterior photograph project along the SAME axis and therefore yield the SAME silhouette outline (mirrored) -- that is ONE shape constraint, not two, however many images you have. Front + LATERAL is what constrains a form. Anterior/posterior pairs are still valuable, but for per-surface ALBEDO, not for shape. ★CSG: the visual hull is an INTERSECTION of back-projected silhouette prisms, and intersection of signed distance fields is max(). Each view is folded in with max, so views can be added incrementally and in any order. Caller supplies cos/sin of the view yaw in Q10 (1024 = 1.0) so this organ needs no trig table and stays exact for the axis-aligned cases that matter (front 1024/0, lateral 0/1024). license_tier: ORIGINAL