Xbox360, so where the hell is the AA?

nAo said:
It's not, for a very simple reason, it would be too slow, c'mon..16x supersampling!
PGR3 does a classic image space motion blur, they render a frame and a velocity buffer which is used in a post process pass to know in which direction a pixel in the frame buffer has to be blurred, that's all.
You can take a variable number of samples, even much more than 16.
Flash fx can be applied after the motion blur pass or even before if it has a zero velocity in the velocity buffer.

I know what a vector blur is. And this is not it.
Take a look at this screenshot.

http://img.gamespot.com/gamespot/images/2005/319/reviews/927245_20051116_screen024.jpg

You can see banding from the passes. A vector blur would not do this kind of banding.
Unless it was very poorly implemented.
And the graphics in PGR are not that heavy, that you couldn't do supersampling.
Granted it might not be 16x, but it sure could be 8x.
 
dubert said:
I know what a vector blur is. And this is not it.
Take a look at this screenshot.

http://img.gamespot.com/gamespot/images/2005/319/reviews/927245_20051116_screen024.jpg

You can see banding from the passes. A vector blur would not do this kind of banding.
Unless it was very poorly implemented.
And the graphics in PGR are not that heavy, that you couldn't do supersampling.
Granted it might not be 16x, but it sure could be 8x.

I'm not so sure about you Dubert. I wanted to believe you yesterday...but you aren't making much of a point here. nAo is a badass in the beyond3d space. I think what he is saying is making more sense. Besides that PGR3 screenshot looks nothing like the 16x supersampling in the ATI demo screenshot you first posted.
 
ROG27 said:
I'm not so sure about you Dubert. I wanted to believe you yesterday...but you aren't making much of a point here. nAo is a badass in the beyond3d space. I think what he is saying is making more sense. Besides that PGR3 screenshot looks nothing like the 16x supersampling in the ATI demo screenshot you first posted.

Check out the demo itself.
http://www.daionet.gr.jp/~masa/rthdribl/

Remember to turn on the effects, and move the view fairly quickly.

EDIT: Here's a link to a better screen capture.
 
Last edited by a moderator:
dubert said:
I know what a vector blur is. And this is not it.
Take a look at this screenshot.

http://img.gamespot.com/gamespot/images/2005/319/reviews/927245_20051116_screen024.jpg

You can see banding from the passes. A vector blur would not do this kind of banding.
Unless it was very poorly implemented.
are you telling me they decided to use a much more espensive blur implementation that gives inferior results instead of another way cheaper techique that could achieve (in your opinion) better results?
It doesn't make any sense.
So they just decided to shade ever pixel 8 or 16 times per frame to have motion blur instead of running the game at a full frame rate?
I don't believe that for a second :) BTW...what a crappy implementation, aliasing is everywhere..:)
And the graphics in PGR are not that heavy, that you couldn't do supersampling.
Granted it might not be 16x, but it sure could be 8x.
Ah well..just 8x..a piece of cake :D
 
Image based motion blur actually has quite a few disadvantages. Objects moving into and out of view, moving along an arc, and moving towards/away from the camera are particularly problematic because there isn't enough information in the rendered image to create a proper effect.
The obvious advantage over any implementation of 3D motion blur is speed (even PRMan is slower, but not that much :))
 
nAo said:
are you telling me they decided to use a much more espensive blur implementation that gives inferior results instead of another way cheaper techique that could achieve (in your opinion) better results?
It doesn't make any sense.
So they just decided to shade ever pixel 8 or 16 times per frame to have motion blur instead of running the game at a full frame rate?
I don't believe that for a second :) BTW...what a crappy implementation, aliasing is everywhere..:)
Ah well..just 8x..a piece of cake :D

Well actually doing it in passes is physically more correct, than doing it in post.
There are artifacts with a vector blur, although it looks more smooth.
Like shadows on a ground plane that does not move, cast from an object that does. The shadows on the ground will not be affected by the vector blur, since the ground has no motion vector. So the shadows will be static, from a moving object.
Other issues are object behind transparent object that have motion. The object behind the transparent object will be affected by the motion vector of the overlaying transparent object.

How else would you explain the banding? A poor quality vector blur?

EDIT: Added 2 example screen shots...
http://www.thefreeimagehosting.com/Uploads/Images/4954014062500vector_blur.jpg
http://www.thefreeimagehosting.com/Uploads/Images/4954143750001supersampled.jpg
The first one demonstrates the inadequacy of a post process vector blur when done on a transparent object. The sphere behind the transparent object is blurred with the transparent object affected by the blur.
The second one is 8x supersampling.

And since PGR3 has the cockpit view with a transparent windshield, that also receives reflections, and dirt textures if you collide with something, that would effectively screw up the post process blur.

Also when did I say it's lower framerate? You can do motion blur to 60 fps just as well. The blur length will just be shorter.
 
Last edited by a moderator:
dubert said:
And the graphics in PGR are not that heavy, that you couldn't do supersampling.
Granted it might not be 16x, but it sure could be 8x.
You're kidding, don't you?
 
Dubert's argument (transparencies and shadows - HDR lighting, too?) is reasonably convincing to me - hmm, maybe that's why the framerate tops out at less than 60fps and why everyone threw their toys out of the pram because it wasn't a guaranteed minimum of 60.

Alternatively, couldn't the game render shadows and transparencies after applying the blur? Though, ahem, erm, that's sorta recursive, argh...

I'm interested in nAo's counter argument...

The photo mode in PGR3 has a variable shutter speed. The game calculates the resulting blur as the shutter speed is varied. Are both of these blur techniques viable in this scenario? I presume they could be.

Anyway, this is a great excuse to watch loads of PGR3 vids to see if I can track what happens with motion blur viewed through multiple windows...

Jawed
 
Laa Yosh said:
Image based motion blur actually has quite a few disadvantages.
Sure, just like Multisampling has a lot of (quality)disadvantages over SSAA. That's why we're all using the latter right? ;)

Anyway the idea of slowing down your renderer by an order of magnitude (he said 16 samples) just to generate motion blur, is completely absurd. That's not a tradeoff, it's a suicide.
 
One thing I've noticed on PGR3 on (old) screenshots and in-game is the fact that reflections are not AA'd.

Or did I miss that discussion?
 
pipo said:
Or did I miss that discussion?

If you did, then so did I. But yeah, playing the game, they definitely don't seem to be AA'd. Wonder if they're in a lower resolution too. BTW, to anyone: reflections are done via rendertargets -> cubemapping, right?
 
TurnDragoZeroV2G said:
If you did, then so did I. But yeah, playing the game, they definitely don't seem to be AA'd. Wonder if they're in a lower resolution too. BTW, to anyone: reflections are done via rendertargets -> cubemapping, right?
They are rendered to offscreen rendertargets which are later used as a cubemap texture yes.
Reflection aliasing artifacts you see in closeups are mostly because they are rendered in fairly low resolution, so they might actually still be using AA.
 
Fafalada said:
Sure, just like Multisampling has a lot of (quality)disadvantages over SSAA. That's why we're all using the latter right? ;)

Anyway the idea of slowing down your renderer by an order of magnitude (he said 16 samples) just to generate motion blur, is completely absurd. That's not a tradeoff, it's a suicide.

Even though the framerate were slower, like 30, you would not notice it because of the motion blur, that's how your eyes work. And who cares if it was 30 fps, since the motion would still look smooth, and have a nice little *real world* property to it, albeit lower quality.
There are issues with a vector blur in a game like this. There are too many transparent obstructions, like the fence in front of the crowd, that has a transparency map. Also the camera flashes in the crowd are transparent quads, which are not blurred, it is present in only 1 pass, when the flash occurs.
How is it *so* impossible that the game uses multisampling? The reflections and alot of other shading properties were removed in splitscreen mode, propably to keep the MB, and the framerate steady.
If it is indeed vector blur, then it is a really poor quality one, since it causes banding that looks like passes, and they would have use multipass tricks to get past the issues with transparent surfaces and shadow motion anyway.
 
dubert said:
EDIT: But where is the AA then?

There is 2xAA in-game (but not in garage). In-game is on 600p native up-scaled on 720p in video output and therefore appearance that there is no aa (in comparison with condemned). Cube-mapping is for reflections; there is no aa on shadows and other cars can not be seen as reflections (I think).
In short, game has graphical problems.
 
Back
Top