KTX textures in model are not loading properly

[jak983464779]

What happened?

Loading a gltf model Some textures do not load properly, and some textures can be displayed

in threejs：

Reproduction steps

...

Sandcastle example

No response

Environment

Browser: chrome
CesiumJS Version: 1.124
Operating System:

[javagl]

A first diagnosis step could be to drag-and-drop one of the models into the glTF validator at https://github.khronos.org/glTF-Validator/ to see whether it contains any errors, and to look at the Developer Console (by hitting F12 in Chrome or FireFox) to see whether it prints any error messages. If there are no errors or messages: Can one of the models be shared? (Maybe publicly, but in doubt, privately?)

[ggetz]

Thanks @javagl for the tip!

@jak983464779 To confirm the model's validity, please run it through the glTF-Validator tool and let us know the results.

[jak983464779]

第一个诊断步骤可能会将其中一个模型拖放到https://github.khronos.org/glTF-Validator/上的glTF验证器中，查看它是否包含任何错误，并查看开发者控制台（在Chrome或FireFox中按F12）可以查看是否打印任何错误消息。如果没有错误或消息：其中一个模型可以共享吗？（也许可以公开，但不确定可以是否私下共享？）

gltf.zip
Hello, this is a GLTF model，Please have a look

[javagl]

The model does generate some error messages. These should be fixed.

But these errors are unrelated to the issue here.

And unfortunately, the issue seems to be somewhat more fundamental. CesiumJS simply does not display some of the KTX textures. I pulled out two textures (one that is displayed, and and that is not displayed) and put them on unit squares:

In the left part, there is a completely black square for that glTF in CesiumJS (below the visible one - a bit hard to see in the screenshot)
The same glTF is displayed properly in https://gltf-viewer.donmccurdy.com/ and https://gltf.report/
The same glTF is not displayed properly in https://sandbox.babylonjs.com/ - it also shows a black square there.

Here are the glTF files and the test sandcastle:

CesiumJS 12397 KTX texture not displayed 2025-01-08.zip

From looking at the basic properties of the KTX files, there was nothing that could explain the different behavior for me:

But interestingly, when running ktx info on both of them, then it did report

• no error for the one that is not displayed
• an error for the one that is displayed

🤪

Specifically, the error one that one that worked is

error-4001: Invalid Level Index. Indices must be sorted from the largest level to the smallest level.
    Indexes for level 8 with byteLength 3 and level 9 with byteLength 4 are incorrectly ordered.
warning-5004: Non-conformant texture file accepted by libktx.
    KTX 2.0 file does not conform to the specification but it is currently accepted by libktx.

The output attached here, just for convenience...

_info_texture_01-broken.txt
_info_texture_02-ok.txt

---

EDIT: One random thought that I had was that it might be the combination of that KTX file having MipMaps, and not being square at the top level. That's not a profound or justified guess, but just ~"the only 'structural' difference'" that I saw between these files...

[jak983464779]

Strangely enough, I modified gltf samplers to display textures normally。

Normal

Abnormal

normal.zip

[javagl]

So there's more evidence that it is related to the MipMapping. I wondered whether there is any connection to #9876 (see the bottom of that PR for some related issues/PRs that came up recently). At least, I could imagine that the fix for this issue may affect a similar area of the code.

EDIT: Just for convenience:

The screenshots suggest that it did not work for minFilter=9987, which is GL_LINEAR_MIPMAP_LINEAR. but it did work for minFilter=9729, which is GL_LINEAR.

The MipMap levels of the texture look fine, though: