fullseye

smallest_box3 — 3D geometry op

使い方

Near-minimum-volume oriented bounding box (the 3-D smallest_rectangle2).

Found by multi-start local refinement: seed the orientation from every convex- hull face normal (the O’Rourke case a candidates — a box face flush with a hull face), from the PCA axes, and from a fixed set of deterministic random frames, then polish each by coordinate descent and keep the least-volume result. This is exact for box-like objects (a rotated cuboid is recovered to machine precision) and, unlike a PCA box (fit_box3 / pcseg.obb), reaches the true minimum on shapes whose optimum has no face flush with a hull face — e.g. a regular tetrahedron, where the PCA / hull-face box is ~2x too large.

Honest limit: this is not a proof of global minimality for every convex shape. The exact guarantee needs O’Rourke’s full case b (two box faces each flush with a hull edge), which is not enumerated here; local refinement drives seeds into that regime instead. Empirically the result is at or below a dense brute-force rotation search, but a pathological shape could leave a small gap.

Returns center (cd/cr/cc), axes (3, 3 — unit ROW vectors), sorted half-extents l1 >= l2 >= l3, full size, volume, and corners (8, 3). Deterministic (fixed random seeds). Raises ValueError on < 4 points or a coplanar/degenerate set (no 3-D hull).

参考(サンプルデータ・文献)

実行できる例(この op を実際に呼ぶ検証済みサンプル)

型が繋がる次の op(primitive を入力に取れる)

fuse_to_voxel · angle_between_lines · angle_between_planes · angle_line_plane · distance_point_plane · distance_point_line · distance_line_line · distance_segment_segment

同カテゴリ(geometry)

line_from_2points · plane_from_3points · angle_3points · angle_between_lines · angle_between_planes · angle_line_plane · distance_point_plane · distance_point_line


Provenance: measure3d.py — 3D operator registry. この per-op ノートは tools/opdocs.py md が自動生成(手編集しない)。

© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.