More WIP on R3PU

[Michael Lobko-Lobanovsky]

I can almost, but not entirely, correct the Ramjet model by using the epsilon of 0.000001, rather than 0.0001, in the (bi)tangent calc. But for the life of me I cannot correct the side planes in Millenium Falcon!

I do not understand your "load/save the model into accutrans" suggestion; there's too little info in it for me. Pure load/save of .OBJ file yields nothing because the file is saved as-is except for the .MAT file. Forcing the AT clone to work with the original .MAT file reveals the glitch. Even if I recalc the hard and/or (not very good) soft normals in AkkuTrans and save them to the clone, using the original .MAT file still reveals the glitch. Which means the model still has some plane normals our (bi)tangents can't cope with!

They can cope only with the (very poor) normals recalculated for the both models by ObjReader itself... 

[Michael Lobko-Lobanovsky]

Tell you what Patrice,

Both Ramjet and Millenium Falcon .OBJ files refuse to be imported in my 3ds Max'es (I have two of them: 32- and 64-bit 2010 and 2016 versions) for re-triangulation. The editors report bad vertex/normal/texture indices. If you exported them as OBJects from your C4D, I regret to say your C4D object exporter is broken.

At the same time, the much smaller test file you had supplied me with earlier does load successfully in 3ds Max and when re-triangulated, exports to a new 1.5 times smaller .OBJ file with perfect normals that both Objector and ObjReader display flawlessly with their existing (bi)tangent calc.

This makes me think that your C4D object exporter is unreliable and may continue to let us down in some future cases without prior notice. What was the original format of these two models? If I can read it with my 3ds Max importers, I suggest we re-triangulate and optimise them first in 3ds Max and then you re-work them again in your C4D to bring them to their ObjReader-renderable state. I think the model files may get down to nearly half their current size in the process.

This way we will also do away with this horror we're seeing which is not our renderers' fault at all.

[Michael Lobko-Lobanovsky]

This is what my 3ds Max did fully automatically with the smaller test file. The two black planes and one trianle at the bottom are backfacing geometry that 3ds Max, as any other 3D editor, cannot correct automagically.

Do compare it against the original smaller test file at the link above and feel the difference.

[Michael Lobko-Lobanovsky]

Were the Ramjet and Millenium Falcon originally provided as C4D format files? As far as I am aware, neither AkkuTrans nor any other 3D editor I know of have C4D proprietary file format importers.

[Patrice Terrier]

Here is a link to the original Ramjet from Anders Lejczak
http://www.colacola.se/expo_alphacentauri.htm

[Patrice Terrier]

To avoid breaking the existing models, i think that the best solution is to keep using the old bump as the default, and use a new mtl option for the new one.

[Michael Lobko-Lobanovsky]

Never! We will never EVER use those sloppy shitty 4-bytes-off-the-target bumps again!!!

We are going to use structurally correct tangent coord offsets and our awesomely cool conventional, parallax, and steep parallax bumps (with optional bent normals) forever in our future work till the day we die!

(And if I ever see this glitch again in some future model, this is going to be a 100% sure indication you just sabotaged your homework... )