GLTF Mesh Import: Model Parts Wrongly Positioned, Scaled and Rotated

[aiaustin]

Environment

Second Life Project Gltf-mesh-import 7.1.14.15122149339 (64bit)
Release Notes

You are at 124.6, 196.7, 23.0 in Rumpus Room 2048 located at simhost-030b947afda1f10ad.aditi
SLURL: secondlife://Aditi/secondlife/Rumpus%20Room%202048/125/197/23
(global coordinates 36,476.6, 14,276.7, 23.0)
Second Life Server 2025-04-24.14648572313
Release Notes

CPU: Intel(R) Xeon(R) CPU E5-1650 v2 @ 3.50GHz (3491.58 MHz)
Memory: 32713 MB
OS Version: Microsoft Windows 10 64-bit (Build 19045.5854)
Graphics Card Vendor: NVIDIA Corporation
Graphics Card: NVIDIA GeForce GTX 980/PCIe/SSE2

Windows Graphics Driver Version: 32.0.15.7652
OpenGL Version: 4.6.0 NVIDIA 576.52

Window size: 1278x889
Font Size Adjustment: 96pt
UI Scaling: 1
Draw distance: 256m
Bandwidth: 2400kbit/s
LOD factor: 3
Render quality: 6
Texture memory: 4096MB
Disk cache: Max size 1638.4 MB (30.1% used)

J2C Decoder Version: KDU v7.10.4
Audio Driver Version: OpenAL, version 1.1 ALSOFT 1.24.2 / OpenAL Community / OpenAL Soft: OpenAL Soft
Dullahan: 1.14.0.202408091639
CEF: 118.4.1+g3dd6078+chromium-118.0.5993.54
Chromium: 118.0.5993.54
LibVLC Version: 3.0.21
Voice Server Version: Not Connected
Packets Lost: 0/34,756 (0.0%)
May 20 2025 03:17:18

Description

Several glTF models after upload have parts which are wrongly positioned, scaled or rotated resulting in the model looking "exploded". This may be due to wrong handling of some transformations in the model. This can lead to extremes of some parts being very large indeed and even limiting the scale at which the model can be imported.

A test model for this might be the GLB Mesh Model of the NASA Mars Ingenuity Helicopter

https://github.com/nasa/NASA-3D-Resources/tree/master/3D%20Models/Ingenuity%20Mars%20Helicopter

Original has textures which are not Power of 2, which may not 9yet) be supported in the mesh uploader (TBC).

SL import of the NASA version of the model (Ingenuity Mars Helicopter.glb) from the NASA web site just indicates "There was a problem uploading the file". SecondLife.log is not very helpful, not even mentioning the model .glb name or the error stronng shown to the user for easy lookup, but I do see:

WARNING # newview/gltf/asset.cpp(498) LL::GLTF::Asset::prep : Unsupported extension:

KHR_draco_mesh_compression

I made a version of the model by importing the NASA .glb into Blender, modifying the textures to be Power of 2, and exporting again to .glb. If a bounding box physics is set this can be imported to SL.

Ingenuity-Po2.zip

But the result is that the parts are out of position, wrongly scaled and/or wrongly rotated. So this is an example of a mesh where parts transforms are not handled correctly yet.

Reproduction steps

1. Look how the Ingenuity-Po2.glb model appears in https://github.khronos.org/glTF-Sample-Viewer-Release/
2. Try to upload Ingenuity-Po2.glb
3. You may need to clear settings and set a physics "bounding box" to avoid "MAV_BLOCK_MISSING" issues or a hang at the "calculate weights and fee" stage.
4. Rez the uploaded model in world
5. Observe the model is "exploded".

[aiaustin]

A rather extreme example of this issue is the attached "Right-Wing.glb" mesh which is a simple 4 part mesh (the wing of a vehicle about 1m in size)...

Wing-Right.zip

This is the model exported from Blender 4.4.3 with the inbuilt glTF exporter as shown in this Khronos glTF Sample Viewer image...

Note that the separate parts do import correctly, but the whole model seems to import as a multi-part object in inventory, rather than the usual single mesh (icon is multi-cubes not the usual single cube) and when rezzed in world has parts that are well apart, rotated and very wrongly scaled. The import has some parts that must be considered (wrongly) by the uploader to be the maximum allowed sized as the max. scale factor shown in the mesh uploader is approx. 0.6 and that cannot be increased in the mesh uploader "Upload options" tab.

This may be a simple example of where transforms are not being applied correctly as this is observed on many meshes where the parts work but the combined mesh doesn't.

[AtlasLinden]

Here's a fun simple one in regards to positioning: https://sketchfab.com/3d-models/springbok-d6a3f63b8bfd4b54a16d260de36fb8b0
And here's what it looks like in the viewer vs Blender:

And also how it appears in the Khronos viewer:

[aiaustin]

Testing Second Life Project Gltf-mesh-import 7.1.14.15192424841 with #4142 commits.

Noted that improvements have been made to the transforms handled. I see @AtlasLinden 's Springbox.glb now does work correctly after import.

Ingenuity-Po2 example uploads but is still exploded on rez.

Wing-Right uploads but does not seem to rez in world (on Beta Grid) though it may be somewhere out of range as its scale was (wrongly) so large.

Please leave this issue open to allow for tests of various models for scale, part positioning and rotation, etc.

[marchcat]

@aiaustin Thank you for checking!

[marchcat]

The issue observed with the Ingenuity-Po2.glb model is only part of the problem, but once resolved, it will likely address most use cases.

[marchcat]

I also use these two models for testing:
https://sketchfab.com/3d-models/toyota-ae86-black-limited-kouki-80854ba302034431b517f523db328129
https://sketchfab.com/3d-models/free-1975-porsche-911-930-turbo-8568d9d14a994b9cae59499f0dbed21e

Before the above commit they both were broken. It's enough to just open them in the model preview floater for now, no need to upload.

[aiaustin]

Good to hear to you take it as intended :-) I tried to find small models that exhibited the range of issues I see, and I agree that once some of these load many others will likely just fall into place.

The old glTF mesh upload/develop preview mechanism worked on the transforms really well up to about June 2024 and then broke and many stopped working. I know its a very different mechanism you are using now, but I used that experience to select meshes I knew showed some transform errors.

I actually have quite a few meshes where the parts upload separately, but not the whole model when parts are combined. But that is likely a different issue.

I am iterating through some test meshes I have set up to try to give feedback, The mesh statistics for those I am using along with Khronos glTF Mesh Preview images of each are at:

https://blog.inf.ed.ac.uk/atate/2025/04/21/second-life-gltf-mesh-uploader-tests/

[marchcat]

It looks much better now.