Forza Motorsport Resource Extraction (.carbin)

[ALYX]

Hey nice work guys. The models in 3ds max are the same smooth as it should. Now only UVMapping remain and texture extractor. Keep up the good work!!!!

[Tosyk]

Hey nice work guys. The models in 3ds max are the same smooth as it should. Now only UVMapping remain and texture extractor. Keep up the good work!!!!

How did you achieve the right results of smoothing in 3d max? I still have all distorted.

[ALYX]

I import model with recompiled Forza Studio with settings in 3ds max 2010 which had been posted on previous page.
Here is result:

[minime891]

When i import into 3ds max it looks ok at fist but when i switch to poly mode you can see the problem.
1st: http://i41.tinypic.com/2r5e2v8.jpg
This is what it is exporting
2nd: http://i39.tinypic.com/35kvi8k.jpg
Imported into zmodeler
3rd: http://i42.tinypic.com/20s7laq.jpg
I can fix it, but is there any way to export it so it looks like the first image?
4th: http://i43.tinypic.com/2002rmg.jpg

Thanks and keep up the fantastic work you guys are doing.

[Dan Frederiksen]

minime, we think it's a vertex position precision problem. the reason it looks good to begin with seems to be that max can use the 'fake' vertex normals that are also used in the game to make the surface appear smooth even though the geometry actually isn't smooth. so what you are seeing to begin with is not actually better geometry. just masked errors.

you said you can fix it. how did you fix it? manually moving single vertices?

[Ernegien]

I still don't believe that this is a precision issue. Each 16-bit float has a 10-bit mantissa with 1/2^10 precision. Since these values are bound between -1 and 1, this cuts the precision in half which equals 0.001953125. Now if it stopped there, yes I would agree this isn't enough precision but these values are each scaled yet again by an equally precise value so when you square that you get an actual precision of 0.000003814697265625 which is more than enough for the purpose they serve. Please correct me if my math is wrong as it's been a while but this seems to be correct on first inspection

After some tests and lots of discussion, Veegie and myself think that the normals may in fact be remnants of quad face data from the original models. When they went to convert their original models to game models, they would break up the model faces into triangle strips. When 3ds max imports this data for the first time, it seems to use the old normals present in the file, but any modification seems to force new normals to be calculated and in some areas they are quite different from the originals which cause these improper smoothing issues...

[revelation]

As far as 3d studio max is concerned, i get the weird polygon structuring as well with 2009. i also think it is a "precision" issue, not with the half float conversion but within the obj importer for 3d studio max. i imported the same obj that give erroneous output into Milkshape 3d an the mesh shows fine, as it does within your application. LOD1 - LOD5 from the main carbin file do not seem to exhibit the problem anywhere near as badly, if present at all. i am guessing either the floating point precision within max during or after the conversion from the text strings within the file could be causing an issue. if it is not that then there is another issue with the obj importer itself as rendering normally in d3d/opengl as your app and milkshape do respectively do not seem to show the problem. As such i highly doubt it is an error in the half float conversion or the text output. But i will continue testing and see what i can find. Expect a pm soon with regard to getting past the "header" information in the file so that the main carbin files can be opened in your program (or tell me if this was already supposed to work, heh).

[Dan Frederiksen]

Ernegien, that's definitely wrong. the precision doesn't improve like that in multiplication. it's a bit complicated to say what the exact step size is because x y and z is not just limited to +-1 they appear to be limited to length 1 of the xyz vector and then scaled. so z y and z are not just free within +-1. but in the simpler case of just one coordinate and a scale both within 1 and say a step size of 0.1 then the extra scale doesn't suddenly allow you to find ten more steps between 0.9 and 1.0. that scaling factor doesn't let you hit 0.98 for instance. roughly speaking I think you can say that the initial stepsize is not improved and even worsened by the scale factor. if we say the precision within 1meter is 2mm (might only be 1 with the implicit extra digit in the float representation, not sure) and then the object is 5 meter in length then the step size is something like 2.5mm-5mm. maybe even as bad as 8mm in some cases.
and you'll mainly see precision issues with large objects.

I believe the precision issue is real. and we can just live with it or try to develop the fancy correcting algorithm.
but I strongly suggest that you all forget about it for now and not bring it up until such time everything else works.
we need the cars extracted as they are, then we might look at making the models even better than they are on the disc

[Ernegien]

Alright, I would have to agree with you there as I didn't entirely think things through. Regardless, I still feel the same as the models are capable of showing up fine in some instances. It is therefore an issue with the normals and or smoothing groups in my opinion.

[Dan Frederiksen]

think of a surface made of legos. it is stepped. but if you clothe the points in normals that can have any direction and interpolate between only the normals then the shading will look as if the surface is smooth. it is not a smoothing group issue.
the vertex normals cloak the precision problem. that's counter intuitive because the vertex normals can actually be something that the geometry itself isn't. it's fake. it works for games. it doesn't work for editors.

[revelation]

[terzer]

what software do you use?

[revelation]

i am just attempting different applications. The screen shots are from Milkshape 3D. 3d studio max shows about the same as other posts here, so i was going to wait until i had side by side LOD comparison shots to be able to post something different from that.

[toolieo]

Tried a few things, although nothing much within changes. I know you guys may know your issue etc, I thought it would be an idea to try all different programs.

So here it goes... Many might say .3ds is crap but hey it was just for testing.
---------------------
3ds Max 2009:


MilkShape: (Thanks revelation for the heads up)


3ds Max 2010:


3ds Max 2010 .3ds Export then loop back import:


Deep Exploration 6 CAD Edition (With DE6's smoothing):


Deep Exploration 6 CAD Edition (Without DE6's smoothing):


Zmodeler (Converted into 3ds from deep exploration 6):

[terzer]

i'm not very good in 3d modeling, i want to ask, i managed to copy only the vertices, without the polygons and the tag, which is creating the smooth surface. is there a way now to generate new polygons?