TetWild producing unexpected results in Prusaslicer

[Sineos]

TetWild:

• Ubuntu 20
• Commit f6ca5d970f151bd57fbec22df9dfe02932e94555
• Called with standard setting: ./TetWild /mnt/d/test.stl

Left is the STL as created by Fusion 360, right the TetWild OBJ file

Various combinations with --stage 2, --filter-energy 7, --is-laplacian did not yield an improvement. STL file is attached (remove txt ending)

test.stl.txt

[Sineos]

I did some more experiments and the results are really strange.

1. Windows, Issue #47
2. Windows, self compiled, commit https://github.com/Yixin-Hu/TetWild/commit/f6ca5d970f151bd57fbec22df9dfe02932e94555
3. Windows, unknown source

The following three images from PrusaSlicer are all created via the command:

TetWild.exe -l 6 -e 0.01 d:\test.stl

Ordering of the images as per the order above:

Attached are the logs of the run with --level 0 tetwild-logs.zip

[Yixin-Hu]

Hi,

What do you mean by "unexpected"? Could you provide both input and output meshes?

For the second question, you are using -e 0.01 which is a quite large envelope size and thus the output surface would be "far" from the input.

[Sineos]

Thanks for coming back to this @Yixin-Hu

What do you mean by "unexpected"?

I tried different combination of parameters including but not limited to:

• default
• -l 1 -e 0.01 --is-laplacian
• -l 3 -e 0.005
• -e 0.025 --targeted-num-v 1000 --is-laplacian
• --stage 2 --filter-energy 7 --is-laplacian

The result was always a more or less severely deformed mesh as shown in the examples above

Now in my eyes the strangest thing:

• I have built the current git under Linux as well as under Windows. Both show this deforming behavior
• A Windows built from the depths of the WWW shows proper results.
  • I would date this build to somewhere around middle 2018 judging from the TetWild -h screen
  • The result is displayed in the 3rd image in my post above

Could you provide both input and output meshes?

The input STL is already attached to the very first post. For the output mesh, which settings would you prefer?

For the second question, you are using -e 0.01 which is a quite large envelope size and thus the output surface would be "far" from the input.

Thanks for the clarification. As mentioned above I tried a lot of combinations already (possible not the right one) and always ended up with the deformed mesh. In addition there is this significant difference to this "unknown TetWild.exe" which is producing a perfect result with -l 6 -e 0.01

[foreachthing]

In addition there is this significant difference to this "unknown TetWild.exe" which is producing a perfect result with -l 6 -e 0.01

I was able to build fTetWild (for Windows) and trying the previously working parameter (as stated above), I come to the conclusion that this version I have is not working correctly (despite the wonderful results)!

Now I can start all over finding working parameters. It would be great if one could enter actual dimensional values, rather than percentages over an unknown diagonal. Example: -l 2mm -e 0.1mm. This would make it easier to find the "correct" parameters - I think.

[Yixin-Hu]

Hi Sineos,

I tried the input using default parameters on MacOS and there is no problem (the output is attached). Is the "severe deformation" you mentioned means that the output surface is different from the input? (TetWild does not exactly preserve the input surface but approsimate it within certain tolerance.)

--is-laplacian flag is needed (bur noe necessary) only for input with open boundaries (e.g. large holes).

Archive.zip