Could activation of SSAA improve texture quality?
- Thread starter Vitall
- Start date
BenSkywalker
Regular
More sharp and detailed? No. With SSAA you are losing contrast, it will not make textures appear more sharp or detailed, the opposite occurs. It can easily make them look better, but not sharper. There were lengthy discussions about this here back in 2K when FSAA was a very hot topic and Rev(amongst others) felt that SSAA degraded texture clarity enough to warrant a negative LOD bias adjustment(which 3dfx provided for users BTW).
BenSkywalker said:
I can't see that I'd agree with that. Using super sampling, you will be making a better approximation of the area that each screen pixel forms when projected onto the texture. This should then give you a much better result. (It also takes into account the occlusion/clipping of portions of the texture)
Errr..., IIRC, the 3dfx approach was to average several 'standard resolution' renders rather than down filter a single high resolution render, and so it would have to have tweaks to the LOD. On other systems, eg PowerVR, Nvidia, ATI, the LOD would not need adjustment.
Sorry, I was trying to be brief for once. 
Figures you wanted the opposite.
As Humus said, Anisotropic Filtering adds sharpness to textures much efficiently. However, using SuperSampling AA can improve on that a bit further in most cases. It is just that the bit extra comes at a significant performance hit, and higher degrees of AF are just a better choice.
For available products: GeForce 4XS mode (this mode specifically, not all modes labelled "XS") is, in my opinion, the best balanced use of SS AA in a modern graphics card. It isn't only SS, though, which is why it is the best balanced.
Some Voodoo cards did better supersampling, and plenty of other cards as well, but at much lower performance levels. Also, the Radeon 9500 and higher cards are technically able to offer a better alternative to the above "4XS" mode (i.e., as balanced but with better output), but they stick to multisampling AA only (doesn't touch textures) and there is no plan for them to change AFAIK (a game developer could do it manually, but that doesn't seem likely either). To offset this, they offer the highest maximum degree of AF (Anisotropic Filtering) commonly available, 16x, and this is their apparent reason for not supporting supersampling.
Of course, some of us would like it if they did both..
Are you looking for a definition of SuperSampling AA? You can search the forums, and look to this (somewhat old) as a guide.
Figures you wanted the opposite.
As Humus said, Anisotropic Filtering adds sharpness to textures much efficiently. However, using SuperSampling AA can improve on that a bit further in most cases. It is just that the bit extra comes at a significant performance hit, and higher degrees of AF are just a better choice.
For available products: GeForce 4XS mode (this mode specifically, not all modes labelled "XS") is, in my opinion, the best balanced use of SS AA in a modern graphics card. It isn't only SS, though, which is why it is the best balanced.
Some Voodoo cards did better supersampling, and plenty of other cards as well, but at much lower performance levels. Also, the Radeon 9500 and higher cards are technically able to offer a better alternative to the above "4XS" mode (i.e., as balanced but with better output), but they stick to multisampling AA only (doesn't touch textures) and there is no plan for them to change AFAIK (a game developer could do it manually, but that doesn't seem likely either). To offset this, they offer the highest maximum degree of AF (Anisotropic Filtering) commonly available, 16x, and this is their apparent reason for not supporting supersampling.
Of course, some of us would like it if they did both.
.Are you looking for a definition of SuperSampling AA? You can search the forums, and look to this (somewhat old) as a guide.
BenSkywalker
Regular
This is the same discussion we had back then
That is more accurate, not more detailed and certainly not sharper.I'm not disagreeing on better, I'm in particular talking about sharper or more detailed.
It wasn't enough at default. Even adjusting the LOD bias to compensate for the comparable resolution that a more traditional SSAA would render out at texture clarity was still reduced considerably. Wish the Basement was up now, I still have several screenshots taken to demonstrate reduced detail and sharpness using SSAA.
Edit
Both; it introduces aliasing while adding visible detail. Sharpness is higher levels of contrast.
Forget about AF. I know what is it, how it work and what it will give me. I dont care. I thought that if i apply SSAA it will somewhat blur my textures(MSAA dont affect textures at all).
Ok. We have 2 screenshots.
On first screen texture is much sharper and ditailed. Could it be result of applying SSAA? Just in theory.
Ok. We have 2 screenshots.
On first screen texture is much sharper and ditailed. Could it be result of applying SSAA? Just in theory.
BenSkywalker
Regular
That is correct. Your screenshots aren't showing up for me.
BenSkywalker
Regular
Post the links instead, they work that way.
I'm thinking SS can represent additional detail, and that the problem with your statement is that it is too universal. Here is my thinking, and point out the errors you see:
I think you have a point in that SS's impact depends on there being more detail to represent that would not be presented otherwise.
Take a shot with no AF...SS is quite simply adding detail in all regards where the texture being sampled has more resolution than its presentation on the screen.
Adding AF, however, it is conceivable that AF may represent all detail already in more circumstance (but not necessarily all), in which case SS on top of that is indeed not adding color detail. But it also seems to me that it is possible that the AF may not represent all the detail of the texture being sampled, which then means that supersampling is indeed adding color detail (this is what I think your comments are missing).
Finally, and entering into my own murky speculation, in cases where it is not providing more color detail, I still think it is providing more positional detail. I think that is what anti-aliasing does.
What I mean by positional detail: higher resolution sampling is represent in the same pixel space, so while contrast may be lost, "positional" error is reduced so that transitions don't "pop" suddenly in motion or manifest as artifacting at edges. The problem with Quincunx (again, to my understanding which I'm asking you to address) is that that it reduces contrast but only adds positional data at the edges.
I think you have a point in that SS's impact depends on there being more detail to represent that would not be presented otherwise.
Take a shot with no AF...SS is quite simply adding detail in all regards where the texture being sampled has more resolution than its presentation on the screen.
Adding AF, however, it is conceivable that AF may represent all detail already in more circumstance (but not necessarily all), in which case SS on top of that is indeed not adding color detail. But it also seems to me that it is possible that the AF may not represent all the detail of the texture being sampled, which then means that supersampling is indeed adding color detail (this is what I think your comments are missing).
Finally, and entering into my own murky speculation, in cases where it is not providing more color detail, I still think it is providing more positional detail. I think that is what anti-aliasing does.
What I mean by positional detail: higher resolution sampling is represent in the same pixel space, so while contrast may be lost, "positional" error is reduced so that transitions don't "pop" suddenly in motion or manifest as artifacting at edges. The problem with Quincunx (again, to my understanding which I'm asking you to address) is that that it reduces contrast but only adds positional data at the edges.
ftp://65.49.78.108/temp/a2.png
ftp://65.49.78.108/temp/a1.png
Someone downloaded this files sucsessfully - but connection is very slow.
ftp://65.49.78.108/temp/a1.png
Someone downloaded this files sucsessfully - but connection is very slow.
BenSkywalker
Regular
Which is why it ends up more accurate.
Adding color values. Think of in terms of bilinear v trilinear. While obviously trilinear has the edge of eliminating mip boundaries, it does not increase the amount of detail beyond elimination of the said artifact despite doubling the number of samples utilized.
Which is a reduction in contrast. Think of it this way. You have a mountain in the distance that's black and white stripes. Backing away from this mountain at some point with current technology that is going to turn gray. With SSAA that will happen quite a bit sooner then without.
That's true, Quincunx is an even larger blur introduction then traditional SSAA.
The problem here is that you are considering detail only as distinct colors.
Consider: when that mountain is black and white striped without SS, you are only evaluating that it has black and white contrasts, not whether the black and white contrasts represent anything other than error. Detail is being lost, even though the idea of contrasts is being preserved.
An extreme example:
You have a 800x600 screen with alternating black and white single pixels in a checkerboard.
You reduce resolution to 200x150.
Without supersampling, you'd represent that as black and white squares 16 times as large. High contrast, but where is the detail?
With supersampling, you'd represent that with grey (depending on the sampling pattern).
Which is closer to the original detail? You'd say that the grey doesn't have distinct colors, I'd say the original image actually looked grey and counting colors is a red herring. I think this applies in less extreme examples as well. The total number of colors was reduced, but the information was not.
By your definition, the first is more detailed, and I think that is not a useful definition of it, because it is quite clearly dropping detail. The same with your black and white striped mountain example.
You maintain that detail is only the number of distinct colors and the contrast between them, not the picture being represented by those colors. I don't see how that is valid.
EDIT: several clarifications.
Consider: when that mountain is black and white striped without SS, you are only evaluating that it has black and white contrasts, not whether the black and white contrasts represent anything other than error. Detail is being lost, even though the idea of contrasts is being preserved.
An extreme example:
You have a 800x600 screen with alternating black and white single pixels in a checkerboard.
You reduce resolution to 200x150.
Without supersampling, you'd represent that as black and white squares 16 times as large. High contrast, but where is the detail?
With supersampling, you'd represent that with grey (depending on the sampling pattern).
Which is closer to the original detail? You'd say that the grey doesn't have distinct colors, I'd say the original image actually looked grey and counting colors is a red herring. I think this applies in less extreme examples as well. The total number of colors was reduced, but the information was not.
By your definition, the first is more detailed, and I think that is not a useful definition of it, because it is quite clearly dropping detail. The same with your black and white striped mountain example.
You maintain that detail is only the number of distinct colors and the contrast between them, not the picture being represented by those colors. I don't see how that is valid.
EDIT: several clarifications.
BenSkywalker
Regular
It isn't my definition of the word
You are talking about an accurate representation of the image, not detailed.Similar threads
