Big difference in behavior between TINGenerator in FME 2021 and 2025

Can you share your sample input? I tried to reproduce what you see by creating a workspace in 2021 and upgrade it to 2025 but I had no different results.

We are in the same boat, running 2021.2.6 and moving to 2025 this year when arcgis enterprise 11.5 is supported.

Hello ​@nielsgerrits,

Thanks for your reply. I work with a subset of the original GeoTiff file. But the subset is still about 85MB in size. So I can’t upload it here. I provide a download link for the geotiff https://zueribox.stadt-zuerich.ch/zueribox/index.php/s/AhNcfBAcdiI8Pxd. I just set up a new blank workspace in FME 2025, read the geotiff, connected it with the TINGenerator, once with Tolerance 5 and once with Tolerance 100. It produced almost the same amount of triangles and more then 1000 times more then in FME 2021.2.6. I really don’t know what’s going on.

Cheers Simon

Ah, I tested differently. Yes I can reproduce what you see, but I’m not sure what to think of it. You probably want to submit a ticket for this one.

​@nielsgerrits thank you for your testing. I think there just a big difference how this TINGenerator works in FME 2025 than it used to work in FME 2021 and I prefer the method used in FME 2021. But maybe I’m missing something here. I will submit a ticket and reference this post here. I hope I can get some hints about it.

Thanks again for your help

Simon

I know that safe were planning on reworking the 3D modelling stuff - possibly this has caused the changes.

If I look at your screenshots I can convince myself that some sort of tiling has been used ​@nielsgerrits  do you see something similar in your testing with different data?

My guess is that FME has added points along the tile boarders and has used these in the triangulation. It’s likely that these boarders are being treated differently when using the surface tolerance - possibly these boarders are treated as breaklines. I’ve seen this in previous algorithms. Have you tried settings the breakline tolerance to anything other than automatic? Try setting it to the save value as Surface Tolerance.    

It’s hard to say if this is a bug or a feature. My guess would be a mixture of the two.

Note: The tiling approach is likely indented to reduce the amount of memory needed by processing the data in chunks. This should also allow for parts of the process to be performed in a multi-threaded approach as well. 

Yes I think I see the tiles forming in my testdata. Must admit I don’t do a lot of work like this, but I can understand the issue Simon has with the different results. And tweaking the surface tolerance parameter does not seem to do very much either.

Thank you guys for your answers and looking in to this. I submitted a request on this topic to Safe and already got a first answer back. Here is what Dave Campanas wrote:

“We did completely re-write the TIN generation code for FME 2024, which is resulting in the differences in the results you are seeing. However, I do agree that the Tolerance is not having the desired effect, so I have created a problem report for our development team (FMEENGINE-86426), requesting another look at the new code.”

So maybe there is an issue with the tolerance setting and they will look into this. Dave also mentioned as a workaround to use a MeshSimplifier after the TINGenerator to simplify the TIN more. I will try this and see what I get out of it. I will let you know, if there are any updates on this topic from Safe.

Thanks again for your help

Simon