Interior Mapping

[Oogst]

Ok, thanks all for participating!

I got a total of 17 test results now, which is enough for the paper. However, I do not have many ATI's in there, so if anyone has an ATI that is not from the X1xxx-series, then I would still like to add that. For the paper, the names will be removed, but I figured it would be nice for checking your own results if the names are there.

http://ronimo.hku.nl/Joost/Testresults.xls

The only really odd numbers there are those from "keHHu n0n3". With the same GPU as a couple of others, he gets twice the framerate. Quite odd and makes me wonder whether the GPU was heavily overclocked, or maybe set to extremely low quality. I don't know, but the result is odd.

A very interesting thing here is that all cards have approximately no change in framerate when the number of rooms is changed. This makes sense, because that is the entire point of the shader. Only three planes are traced per pixel, no matter how many rooms there are. However, the ATI cards do show a difference, up to even 10%! I can imagine why: more rooms means that the texture coordinates of the walls will vary more among neighbouring pixels (there is a jump every time a new room starts) so caching would go wrong there. Still, it surprises me that the other cards do not notice this, not even some other ATI's.

The next steps for me now are:
-finish the paper
-improve the visual quality of the textures
-send the paper in for Eurographics in Kreta (Wednesday next week)
-post the paper here
-create a nice demo with actual art and a little street with buildings with interior mapping
-hear whether it was accepted by Eurographics (end of November)
-if not accepted, send it in for CGI Conference in Istanbul (deadline December 1st)
-continue working on the rest of my Computer Science Master's thesis, which is about raytracing in pixel shaders in general (Interior Mapping is just the first example)
-hopefully, present the paper in April at Eurographics in Kreta or in June at CGI Conference in Istanbul

[Fiesch]

great work and good aproach good luck with your paper!

[kenny.bsp]

The only really odd numbers there are those from "keHHu n0n3"

That would be me

whether the GPU was heavily overclocked, or maybe set to extremely low quality.

No verclocking or low quality settings at all ... Everything was set up as said in the .txt and driver is set to let the App decide the settings.

[REiNDEeR]

muhaha I knew my system would turn out most crappy, nice to see the results

[Oogst]

The only really odd numbers there are those from "keHHu n0n3"

That would be me

whether the GPU was heavily overclocked, or maybe set to extremely low quality.

No verclocking or low quality settings at all ... Everything was set up as said in the .txt and driver is set to let the App decide the settings.

Do you have SLI then? I really odn't understand these results.

[matches81]

Those results look really promising. I am looking forward to the paper, too.
However I have a problem running the benchmark app on my system. Right after start I get a RenderingAPIException:

OGRE EXCEPTION(3:RenderingAPIException): Failed to DrawPrimitive : An undetermined error occurred in D3D9RenderSystem::_render at \Projects\OgreCVS\Branches\Eihort_vc71\ogrenew\RenderSystems\Direct3D9\src\OgreD3D9RenderSystem.cpp (line 2719)

I have a Core 2 Duo E6600, 2GB RAM and a GeForce 8800 GTX, running the latest Forceware drivers, and I chose Direct3D9 as rendering subsystem and all the other settings suggested in the explanations.txt file.

The log also contains a warning about box.mesh being an older mesh format.

[kenny.bsp]

Do you have SLI then? I really odn't understand these results.

No SLI...
I think I found the problem ... nvidia driver that I installed a few weeks ago doesn't change the image settings... If I set them to quality with the new driver it kills my fps (well the results are way smaller then they were)
Guess you should remove me from the result doc :d
It probably was these settings that was on performance all the time no matter what I was setting..

[Oogst]

Those results look really promising. I am looking forward to the paper, too.
However I have a problem running the benchmark app on my system. Right after start I get a RenderingAPIException:

OGRE EXCEPTION(3:RenderingAPIException): Failed to DrawPrimitive : An undetermined error occurred in D3D9RenderSystem::_render at \Projects\OgreCVS\Branches\Eihort_vc71\ogrenew\RenderSystems\Direct3D9\src\OgreD3D9RenderSystem.cpp (line 2719)

I have a Core 2 Duo E6600, 2GB RAM and a GeForce 8800 GTX, running the latest Forceware drivers, and I chose Direct3D9 as rendering subsystem and all the other settings suggested in the explanations.txt file.

The log also contains a warning about box.mesh being an older mesh format.

Often a reboot solves that problem, or not running 3dsmax with 2 million polygons in the viewport while starting the game. It is an irritating problem, because it is totally unclear what causes it. We had things like that with De Blob too.

Do you have SLI then? I really odn't understand these results.

No SLI...
I think I found the problem ... nvidia driver that I installed a few weeks ago doesn't change the image settings... If I set them to quality with the new driver it kills my fps (well the results are way smaller then they were)
Guess you should remove me from the result doc :d
It probably was these settings that was on performance all the time no matter what I was setting..

Ah, ok, thanks for looking into it! I will remove your scores from the results.

[Oogst]

Okay, I send in my paper to Eurographics last Wednesday, so now I just have to wait and hope... The selections will be announced half November.

I made some better textures than what I had so far for my paper for Eurographics and although I had to rush them like crazy to get it done in time and am not totally happy with them, the whole thing does look a lot better with these at least slightly sensible textures, making it look a bit less like Wolfenstein.

The latest version with these new textures can be found here:

Interior Mapping version 10

This is a little video I made of the thing:

Interior Mapping v10 video

And these are some new screenshots:









Now the main thing I have left to do, is to make a little street-scene that shows of the effect in context, so actually looking good graphically and not looking like a box with a black background. I have a little question about that: would Ogre be interested in using that as a new Ogre-demo, if I made it nicely? It would show both Interior Mapping and lightmaps and an animated camera.

[PolyVox]

Very nice! The new textures look great. Good luck with the paper - EuroGraphics is tough but I think you have a good chance. Btw, you included the source in that latest .rar file. Did you mean to? It wasn't in earlier ones.

[Oogst]

The source was in most RAR-files, I only left it out in the benchmarking versions. I deliberately put it there, because research is supposed to be open, so I figured the source should be open to have a look at too.

[Kencho]

Dude, I love you We're in the starting process yet with our graduation project, but this will sure be the perfect companion once we have something decent to show.

[PolyVox]

The source was in most RAR-files, I only left it out in the benchmarking versions. I deliberately put it there, because research is supposed to be open, so I figured the source should be open to have a look at too.

I completely agree, but the benchmarking version was the only one I looked at previously so I thought maybe you'd included the source by accident. Is it under any particular license?

[mr.Zog]

Looks really great

Good luck!

[Lothar]

One word - Outstanding

[Lee04]

Very good work.

Have you tried adding a faint environment reflexion component to make the windows look a bit more like glass?

[ppClarity]

Very impressive.

A small bug you've probably noticed:

Code: Select all

+-----+A
|     |
|     |
|     |
+-----+
C      B

         _|
           camera

Assume that all corner rooms have two windows. The B rooms correctly render both windows, but the A and C corners render their far walls.

[Oogst]

Very good work.

Have you tried adding a faint environment reflexion component to make the windows look a bit more like glass?

Actually, it is in all the screenshots already, it is just that I set it to be too subtle to be noticed in screenshots. If the camera moves, then the reflection maps are very clear. In actual games, the reflection would of course have to be much stronger and the interior much weaker.

Dude, I love you We're in the starting process yet with our graduation project, but this will sure be the perfect companion once we have something decent to show.

It would be great if people used it! Please let me know if you are actually using it and have something to show!

... Is it under any particular license?

Nope. Just the license that you should be nice and not pretend its your own work , but for the rest it is free to use. And I would like to know if anyone used it, just because it is fun to know!

Very impressive.

A small bug you've probably noticed:

Code: Select all

+-----+A
|     |
|     |
|     |
+-----+
C      B

         _|
           camera

Assume that all corner rooms have two windows. The B rooms correctly render both windows, but the A and C corners render their far walls.

Yes, that is exactly why the thing can be done so efficiently. It gets even better if the walls are positioned badly: the far wall will be far beyond the end of the building, which is quite odd... Interior Mapping is efficient because it does not need to know anything about the further geometry, I do not see a solution for this problem that will still yield acceptable framerates. If you have a suggestion to solve this, please let me know. By the way, you have not yet noticed how wrong things go at corners or curves in buildings if you turn in a material with alpha planes. Try the sphere with alpha planes and you will notice it is not geometrically correct at all. Looks good on normal buildings, though.

[jjp]

By the way: Half Life 2 probably is a game where this technique could improve the look. In the city environment there are lots of buildings that have non-transparent windows with environment reflection. I guess slightly visible interiour maps would help the credibility of these scenes a lot (actually the non-transparent windows were among the most common complaints about the graphics in that game).

[Kencho]

Dude, I love you We're in the starting process yet with our graduation project, but this will sure be the perfect companion once we have something decent to show.

It would be great if people used it! Please let me know if you are actually using it and have something to show!

Not right now, but you can be sure you'll know when we use it

[Oogst]

By the way: Half Life 2 probably is a game where this technique could improve the look. In the city environment there are lots of buildings that have non-transparent windows with environment reflection. I guess slightly visible interiour maps would help the credibility of these scenes a lot (actually the non-transparent windows were among the most common complaints about the graphics in that game).

I had not thought of that game yet for this, so far I only thought of games like Spiderman, Superman, Crackdown and Grand Theft Auto, but I agree that Half-Life 2 would greatly benefit from this technology.

[Aladrin]

At the beginning, this technique was extremely efficient... Now that you've added so many features, is it still very efficient, or has it come closer to just rendering the stuff as normal?

[Oogst]

Depends on the size of the city and the features used. Interior Mapping with rooms and four different textures (two for walls, one for floors and one for ceilings) is what I consider to be the basic choice and is definitely fast enough for usage in games. It is faster than real interiors as well, because these just cost a lot of polygons and render calls. Using four different tiling textures on polygonised interiors is a whole lot of render calls. Interior Mapping quickly wins in that case. My paper contains a more detailled comparison of who wins, but I think Interior Mapping is definitely worth it performance-wise, although for a game with such small environments as Half-Life 2 I am not sure whether it is better than real interiors.

Development-wise, Interior Mapping is definitely a winner, because you do not need to model anything. The only assets needed are a few interior textures and that is just a great benefit to development.

[nedelman]

This is a really great looking effect. Excellent work. And in case you're curious, the demo runs between 600 and 800 fps on my NVidia 7900 GT.

Have you played Crackdown on Xbox 360? It has a similar effect, though it's used on just the floors and ceilings. It definitely adds an extra bit of realism to the scene though.

[Oogst]

A friend of min had a 360 and analysed the Crackdown system. They are doing something different then what I am doing, although with similar effect. Indeed only floors and ceilings and they fade them away in the distance, like heavy fog. However, they calculate them such that the orientation of texture on the floors and ceilings depends on the angle of the polygon, so they are doing something different.

Anyway: you are right, they are doing something similar, but simpler.