Future of MSAA?
- Thread starter tEd
- Start date
It's not just the textures. Quincunx (and 4x9tap) will "smoothen" all edges, always. This obviously includes edges that are already exactly representable and thus don't need smoothing. "Smoothing" per se is not a desirable effect. It destroys information. This is why people with bad eyes get glasses.
With Quincunx it's impossible to display a white axis aligned rectangle on a black background without getting a one pixel wide grey halo around it. I just don't get how this is supposed to be a quality improvement.
With Quincunx it's impossible to display a white axis aligned rectangle on a black background without getting a one pixel wide grey halo around it. I just don't get how this is supposed to be a quality improvement.
radar1200gs
Regular
Xmas said:
I can assure you that they do. The farcry image was generated by the game itself (hence the .jpg format).
The nolf2 image was obtained by pressing printscreen, dumping the resulting image into irfanview and saving unmodified as .png.
Zekensack, the blur filter will, like all blur filters make a mess of one pixel wide high contrast lines etc. I haven't found it to be much of a problem (it's a side-effect I can live with, and if I do find it objectional for a particular game I can simply use a different form of AA).
The combining of samples from other pixels is done within the just before / at the RAMDAC - the image is never stored like this, so there's no guarantee that using Printscreen actually works.
radar1200gs
Regular
The drivers are supposed to take account of what happens in the ramdac a screenshot is requested or don't you remember the mini-controversy over this when NV3x was first introduced? EDIT: I wouldn't be able to demonstrate the blur on quincunx if the drivers didn't take account of it.
I believe that gamma values for FSAA are hardwired int r3xx/R4xx hardware.
I believe that gamma values for FSAA are hardwired int r3xx/R4xx hardware.
radar1200gs said:
Do you just selectively not read things? The very thread where this was discussed mentions that it's set to the game gamma settings.
radar1200gs
Regular
Well I guess if you like throwing gamma for the entire scene out just to correct an AA issue doesn't bother you that might be a solution.
And I don't believe you can adequately compenstate with brightness/contrast adjustment either. If they could acheive what gamma correction does games (and other software) wouldn't have gamma adjusters.
And I don't believe you can adequately compenstate with brightness/contrast adjustment either. If they could acheive what gamma correction does games (and other software) wouldn't have gamma adjusters.
radar1200gs said:
What??? I can only assume you have little capabilities for reading or comprehension.
It take the default gamma ramp into account under default conditions, but if you alter the gamma in the game then the gamma ramp you apply in the game are used.
radar1200gs
Regular
You are darkening the entire scene (whether its an edge or not) in an attempt to correct gamma corrected edges. (ATI's gamma correction should only affect the edges as OpenGL guy said in another thread, because it is MSAA it won't touch the texture data except where it lies on an object edge).
That makes no sense whatsoever to me.
That makes no sense whatsoever to me.
You are not "darkening the entire scene".
When you are displaying an image it will have a gamma ramp applied to it (by default) to best best suite the output - this always occurs, FSAA or not. You can opt to tweak that ramp for your desktop or in individual games you get the image further to your liking on your particulr monitor.
With FSAA, because the gamma ramp is not linear, calculating the subsample weightings without accounting for gamma can result in averaged pixel colour that is not necessarily the best for the gamma ramp of your monitor; if, however you take into account the gamma ramp as a part of the weighting for the subsampe values, the resultant pixel colour value should be more appropriate for your display.
You have to bear in mind that the gamma ramp is there all the time - your normal desktop display, non-FSAA rendering, or FSAA rendering. All ATI's gamma corrected AA is doing is taking into account the gamma ramp that has to be there when weighting the subsample values to get a more appropriate final pixel for your monitor.
When you are displaying an image it will have a gamma ramp applied to it (by default) to best best suite the output - this always occurs, FSAA or not. You can opt to tweak that ramp for your desktop or in individual games you get the image further to your liking on your particulr monitor.
With FSAA, because the gamma ramp is not linear, calculating the subsample weightings without accounting for gamma can result in averaged pixel colour that is not necessarily the best for the gamma ramp of your monitor; if, however you take into account the gamma ramp as a part of the weighting for the subsampe values, the resultant pixel colour value should be more appropriate for your display.
You have to bear in mind that the gamma ramp is there all the time - your normal desktop display, non-FSAA rendering, or FSAA rendering. All ATI's gamma corrected AA is doing is taking into account the gamma ramp that has to be there when weighting the subsample values to get a more appropriate final pixel for your monitor.
I can prove you that they don't.radar1200gs said:
Just look at the trees. The alpha test leaves show sharp contrasts from one pixel to the next. This is impossible to achieve with a blur filter like Quincunx.
Well, maybe the hardware is actually able to do this. But if that's the case ATI might want to prove it and eventually implement it in their drivers...DaveBaumann said:
It would be automatic in the first place...
radar1200gs
Regular
Go ahead and prove it then xmas. I have posted screenshots exactly as generated.
Daves whole gamma argument is ridiculous. No-one would be able to do any precision color work whatsoever on a PC if you were to believe him. And its an extremely simply matter to post a few screenshots showing the effects of default, lowered and raised gamma on a game.
Daves whole gamma argument is ridiculous. No-one would be able to do any precision color work whatsoever on a PC if you were to believe him. And its an extremely simply matter to post a few screenshots showing the effects of default, lowered and raised gamma on a game.
Oh, come on...radar1200gs said:
View the FarCry shot with a program that can zoom images without bilinear upscaling, unlike the WinXP image viewer. Take the upper left corner, zoom in 4x or so. There's the edge of a dark object, and the alpha tested leaves of a tree. You see very bright pixels directly neighboring very dark ones, right? Do you see any blur there? No? Well, that's because there's no blur filter in effect.
The maximum difference between two neighboring pixels is 192 (per color channel) if Quincunx or 4x 9tap is enabled. The reason for this can be found in the downfilter kernels I posted: each pixel shares a quarter of the samples with each neighbor.
I can easily find neighboring pixels with a higher difference, e.g. (1,132) and (2,132) with an RGB value of (6, 17, 11) and (208, 222, 233), respectively. This doesn't happen as an effect of JPEG compression, so the only possible explanation is: there is no blur.
Dave is correct in that gamma has to be taken into account for any kind of color blending in a non-linear color space. The problem is, most applications, esp. games, give a damn about linearity of color space.
radar1200gs
Regular
You'll have to take it up with someone at nVidia, xmas. Those screenshots an an accurate representation of what I see in game.
I'm not sure exactly how 4x9 Tap AA actually works. It could well be nVidia is smart enough to turn the blur off when there is no need for it.
On some OpenGL games you can see more blur (mainly on in game text where it very much resembles 2xQ). Why there is a difference between D3D and OpenGL I do not know.
I'm not sure exactly how 4x9 Tap AA actually works. It could well be nVidia is smart enough to turn the blur off when there is no need for it.
On some OpenGL games you can see more blur (mainly on in game text where it very much resembles 2xQ). Why there is a difference between D3D and OpenGL I do not know.
radar1200gs said:
Then clearly you don't have any idea about how images are actually displayed. This is not "my argument" this is how things work!
