Repository navigation
Serpentine Gallery
Each sample Rhino .3dm in the openNURBS V1–V6 set is imported with
import3DM (the Import Gallery thumbnails) and, on the
same page, rendered independently by Serpentine3D so the two
interpretations of the file can be eyeballed side by side. Two different
kernels (FreeCAD/OCCT vs Serpentine3D) reading the same file should agree on
what is visible, which makes this a useful two-way check on import fidelity.
Updated for Serpentine3D 0.10.7 and the import3DM trimmed-face fix (October 2026): all thumbnails — both sides — were re-captured with Serpentine3D 0.10.7 and the fixed import3DM. The original run was made against Serpentine3D 0.9.0 and is preserved at Serpentine Gallery (0.9.0).
A few representative rows:
![]() import3DM — v1_Camera |
![]() Serpentine3D — v1_Camera |
![]() import3DM — v1_HumanHead1 |
![]() Serpentine3D — v1_HumanHead1 |
![]() import3DM — v1_Soccer |
![]() Serpentine3D — v1_Soccer |
![]() import3DM — v4_TreeFrog |
![]() Serpentine3D — v4_TreeFrog |
The 0.9.0 run found two faults, both since fixed upstream — Serpentine3D issue #10 (closed):
-
Six files hung or crashed Serpentine's
.3dmimporter —v4_Gear,v4_RhinoPhone,v4_Wheel_PG,v5_ring,v5_clip,v5_disk_brake. The cause was extrusion faces with a seam, walked twice so the boundary never closed; the importer now uses the trim loops the file already carries (0.10.2). All six import on 0.10.3 — verified individually. - "Needs a fresh start" — a Rhino point object had no importer branch, so a hidden point left a shapeless object that then poisoned every later read of the scene. Fixed in 0.10.3.
Thumbnails only catch what is visible, so each sample was also imported by
both programs and the total surface area compared — and, where the file
stores them, checked face by face against the render meshes Rhino saved in
the .3dm, an independent third reference.
The first run (Serpentine3D 0.10.3) flagged 17 files where Serpentine read 5–89 % less area than import3DM, reported as issue #34 (closed). It turned out to be faults on both sides:
- Serpentine3D — faces running to a pole or round a seam were rebuilt on the wrong side, and watertight shells were left open. Fixed in 0.10.5 (current 0.10.7): TreeFrog now imports as one closed solid with its eye domes, and rhino_logo, MatchSrf and DinerMug recover their area. Serpentine also now splits creased (C0) faces before measuring area.
-
import3DM — most of the remaining "shortfall" was import3DM
over-reading. When a trimmed face could not be rebuilt it fell back to
the whole untrimmed surface, and holes (inner trim loops) were lost.
Causes and fixes:
- the size guard compared the trim wire with
BoundBox, which on a trimmed B-spline face is the box of the surface poles, so correctly trimmed faces were rejected — it now uses the tight geometric box; - holes are cut reliably (all at once when valid, otherwise one at a time);
- loops that walk a seam twice or run to a pole are rebuilt in the surface's own (u, v) — the same class of fault Serpentine fixed;
- no face may exceed its untrimmed surface's area, and the analytic cone fit is validated (it was building the mirror-image cone).
- the size guard compared the trim wire with
Across the 91 sample files with Breps, import3DM now builds every face (none failed, none invalid), and no face falls back to an untrimmed surface except where the file's own trim cannot be rebuilt.
The import3DM figures here were measured on its native trim path
(parse_rhino_breps, no trim3dm build), so they can be reproduced without
compiling anything. With trim3dm built, current import3DM gives the same totals
(e.g. v4_Wheel_PG within 0.05 %); see the note on
Trimmed Surfaces about an earlier trim3dm-path fault that
did not affect this cross-check.
| File | Serpentine3D 0.10.7 | import3DM (fixed) | Rhino render mesh | import3DM before fix |
|---|---|---|---|---|
| v4_TreeFrog | 6.672 | 6.671 | 6.66 | 6.658 |
| v4_WishBone | 1,077.3 | 1,076.6 | 1,078.0 | 1,625.8 |
| v4_SaltAndPepper | 26,053 | 26,106 | 26,054 | 29,521 |
| v5_disk_brake | 7,051.6 | 7,056.8 | 7,029.2 | 9,746.4 |
| v5_rhino_logo | 1,058.2 | 1,057.5 | 1,056.7 | 1,105.3 |
| v1_MatchSrf | 33,404 | 33,454 | — | 34,445 |
| v4_Wheel_PG | 2,565,262 | 2,569,542 | — | 2,913,238 |
| v1_T-Joint2 | 2,253.2 | 2,257.1 | — | 2,467.4 |
| v4_DinerMug | 42,281 | 34,266 * | 42,256 | 35,458 |
* DinerMug's body is one creased surface. FreeCAD's Shape.Area uses
OpenCASCADE's default integration, which under-reads across creases (the body
measures 30,850 against 38,845 in Rhino's mesh); the face itself is trimmed
correctly. Serpentine splits such faces at their creases before integrating.
"—" means the file stores no render meshes. Where it does, both importers now agree with Rhino to within about 0.5 % (DinerMug apart, as above).
Full sweep, Serpentine3D 0.10.7 vs fixed import3DM (138 sample files):
| result | files |
|---|---|
| agree within 5 % | 91 |
| Serpentine lower | 0 (was 17) |
| Serpentine higher | 9 |
| no surfaces in either | 25 |
| surfaces only in Serpentine | 13 |
Every #34 file agrees to within 0.2 %. The "higher" and "only in Serpentine"
rows measure different things rather than faults: Serpentine's total includes
mesh and SubD area, while the import3DM sweep sums only Part::Feature
surfaces (SketchOnMesh, Soccer, example_file, the my_mesh, Flow3, dril and
SubD-logo samples) — plus DinerMug's crease measurement above.
The full per-file table and the sweep scripts are in the repository under
tools/serp_gallery.
Serpentine3D 0.9.0 ignored Rhino layer visibility — every layer arrived visible — so the 0.9.0 captures hid the Rhino-invisible layers via the layers RPC after import.
0.10.3 honours visibility: files with hidden Rhino objects (for example
v4_Wheel_PG's inner parts) import as hidden, matching the file's saved view,
so that post-import layer-hiding workaround is no longer needed.
Running both importers against the same files has surfaced faults on each side — which is the point of the gallery:
- Serpentine3D — the six hanging files and the scene-poisoning point (0.10.x), and faces at poles and seams losing area (0.10.5, #34).
-
import3DM — hidden-layer construction surfaces no longer leak into
imports; open-profile extrusions import as surfaces; annular (pipe)
extrusions now import hollow instead of filled — an outer-plus-inner-circle
extrusion was being solidified into a plain cylinder (spotted on
v5_extrusions); and trimmed faces no longer fall back to the untrimmed surface or lose their holes (spotted through the #34 area cross-check — WishBone, SaltAndPepper, Wheel_PG, disk_brake, T-Joint2, RhinoPhone, clip).
The complete comparison — 138 rows, one per sample — is generated into the
repository at
serpentine_gallery/index.html
(open the downloaded index.html in a browser; the two image folders sit next
to it). The 0.9.0 run captured 132 of 138 — the six above showed a "not
captured" cell. Re-captured against 0.10.7 with the fixed import3DM, all
138 capture on both sides with no errors.







