Why did the software fail to identify the self-intersecting triangular faces?

[jiushi-ai]

Why didn't the software identify the self-intersecting triangles?

After multiple iterations of repairing a problematic model using Mesblib, I found that most of the issues were resolved. (Under these parameters, the model was detected as very healthy, with 0 errors.)
However, when I used Magic-RP, I found a clear error: four triangles were self-intersecting. But the MeshInspector software failed to identify or handle this issue.
Magic-RP detection results are shown in the image:

MeshInspector detection results are shown in the image:

What could be the reason why MeshInspector failed to identify this self-intersection problem, which appears to be a clearly erroneous area?

self-inter-error.zip

Please help me investigate the reason.

[Grantim]

Hello!

The issue here that these faces are overlapping (which we differ from intersecting). But with "Touches as Self-Intersections" they should still be detected.

Current implementation contains a bug and misses this case, we will fix it for next release, thanks for reporting!
(the reason is that we detected only simple touches instead in-plane intersections)

[jiushi-ai]

Thank you for your attention

I also enabled the "Touches as Self-Intersections" option, but the software still didn't detect this issue.

The problem area consists of four triangles. Three triangles overlap on the same face, but one triangle is not on the same face as the other triangles; it is perpendicular to the other triangles and intersects with them. So strictly speaking,

these four triangles have both partially overlapping faces and intersecting areas.

[Fedr]

We updated self-intersection detection in the master branch of MeshLib, and now these triangles are detected (if touches are considered as self-intersections):

[jiushi-ai]

OK, I'll update the lib and test it. Thanks for your follow-up.

[jiushi-ai]

My error detection model identified two self-intersecting faces, but I still have two questions:

1. My local Python library version is meshlib 3.1.0.75 (this is a library version prior to February 1, 2026, but it still detected two self-intersecting faces in today's test. I haven't updated my local meshlib version to Release v3.1.1.92; version 3.1.0.75 detected them, which is strange, as it should only be detected with the latest library).

My local version is shown in the image.

2. Referring to Magic RP, it detected four faces (including the two detected by our meshlib).

See the image below.

this is meshlib image：

[Fedr]

My local Python library version is meshlib 3.1.0.75 (this is a library version prior to February 1, 2026, but it still detected two self-intersecting faces in today's test. I haven't updated my local meshlib version to Release v3.1.1.92; version 3.1.0.75 detected them, which is strange, as it should only be detected with the latest library).

Yes, the improvement that allowed us to detect these two triangles was already in meshlib 3.1.0.75.

Referring to Magic RP, it detected four faces (including the two detected by our meshlib).

It looks like that software detects not only intersecting or touching triangles, but also the triangles that are closer one to another than some threshold (if you believe that we do not report some really intersecting triangles, please give us an example from two triangles only). At this moment we do not have plans to detect close triangles.