Trilinear Test (Warning 0.5MB PNG)

[Bambers]

sigh double post, :|

 

[crushinator]

As you may have noticed, I did choose right. That was just because I felt the right side was slightly more blured or left side was slightly sharper. I watched the picture only 30 seconds, not using any analysing tools, but knowing that traditional trilinear filtering could overblur, guessing what "trylinear" might do, using a 21" Trinitron CRT (0.25 mm pitch)

Note, I wasn't sure my choice was correct, but I was sure I'd like the left side more, since it seemed to be sharper. Subjectlively "trylinear" might look better to me than that traditional one. But let's judge after seeing "trylinear" in motion.

 

[Pete]

Bias meter: Single Post [x----|-----] Double Post

 

[mjtdevries]

Dave,

You might want to post another of these polls.

But the next time people should just use their eyes and not special tools.
And people should not post what they think is the correct one. That way votes won't be influenced by the replies.

And the poll should ask which side looks best. After all, that is more important than which technology is used.

I hope there will be several more of these blind tests.

 

[vb]

so, everybody is choosing right because left is crisper and more prone to moire effects

no one has seen any mip map banding

I have a solution:

Lets call this Quadrilinear ®.

 

[KimB]

Since we don't know what the technique is doing, it's kinda hard to judge what image quality artifacts we should be looking for, isn't it?

 

[mjtdevries]

To judge IQ you don't need to know what to look for.

If you want to find worst case scenario's then it can be usefull.

 

[KimB]

To judge IQ you don't need to know what to look for.

If you want to find worst case scenario's then it can be usefull.

Yes, you do. You just can't look at every possible in-game scenario in any limited time. While I'm sure the shortcomings of ATI's technique will come out eventually, the simple fact is that this technique is against ATI's own claims about what can be considered, "acceptable optimizations," which, in ATI's documents, does not include anything that does less work.

 

[mjtdevries]

No you don't.

I'll grant you that it is usefull if you want to determine the worst case scenarios and what to use that knowledge to determine if it will be visible in games, without testing each and every one of them.

But in this case where you have do a blind test on a single screenshot (or a small number of tests) to judge the IQ, you don't need that info.

 

[KimB]

But in this case where you have do a blind test on a single screenshot (or a small number of tests) to judge the IQ, you don't need that info.

No. First of all, nobody plays still frames. No matter the game, you're going to be moving. Different artifacts become visible in motion than you see when still. Game screenshots, then, are generally a poor indicator of image quality. Better indicators are synthetic programs that are designed to look at one specific aspect of image quality, an aspect that will be more noticeable in real-game scenarios than a static screenshot, where you're actually moving.

Furthermore, I am of the opinion that better image quality means removing of artifacts, whether that means aliasing, blurring, or discoloration. The better technique is not the one that looks better in situation X, but the one that has a better worst-case scenario, and is therefore less likely to become a distraction.

 

[mjtdevries]

I can agree with that.

A blind test with a moving game scenario would be fantastic.

Also that the better technique should have the best worst case scenario is acceptable. (although you can wonder if that applies well to adaptive AF. You might consider a trade-off worhtwhile even though it has a worse worst case scenario)

But in both case you still don't need to know what to look for to judge IQ. For a good objective IQ judgement, you shouldn't know what to look for.

In fact, we are in this mess precisely because people thought they knew what to look for. They thought it was enough to look at coloured mipmaps instead of judging the end result on screen with their bare eyes.

 

[KimB]

(although you can wonder if that applies well to adaptive AF. You might consider a trade-off worhtwhile even though it has a worse worst case scenario)

I definitely don't think so. The off-angle problems of the R3xx mean that a number of my games currently, game that very frequently do not have flat terrain, become blurry in places. This was most noticeable in Morrowind (due to the slow pace of the game), and Everquest (where aliasing reared its ugly head).

In fact, we are in this mess precisely because people thought they knew what to look for. They thought it was enough to look at coloured mipmaps instead of judging the end result on screen with their bare eyes.

Um. The colored MIP maps look fine, because the technique is disabled when using them.

 

[Quitch]

The technique isn't disabled for coloured mip maps, it decides that full trilinear is required due to the colour differences.

 

[KimB]

The technique isn't disabled for coloured mip maps, it decides that full trilinear is required due to the colour differences.

This decision is made at texture load time. I don't see the difference.

 

[mjtdevries]

What I meant with the coloured mipmaps remarkt was that people were judging IQ by looking at the filtering technique used instead of looking at the image quality itself

They thought that if the mipmaps showed proper trilinear filtering that that also meant that IQ is high. But that is not a proper way to test IQ. Usage of some algoritme does not necessary equal high IQ or even best possible IQ.

IMO the real issue here is: what is the definition of high IQ? Is it:
- highest IQ == usage of algoritmes A and B
or
- highest IQ == whatever looks best to the human eye

The first option is what review sites have done so far. Check if full trinlinear filtering is used. If yes, then that means that IQ must be good.
They would also love to do a bitwise comparison with a reference render to give a numerical value for IQ. And that also makes it simple to declare a "winner".

The second option means you have to do a doubleblind test with multiple test subjects. That takes more time, is more difficult to explain to your users, and results cannot exactly be reproduced.
(not to mention that biased review sites can easily mess with the results afterwards)

 

[bloodbob]

The technique isn't disabled for coloured mip maps, it decides that full trilinear is required due to the colour differences.

No it doesn't check it all they check is if it is auto-generated with a box filter. If it is optimise like you would on your girlfriend with a hot chick. If it isn't don't optimise.

Now I'd like to know how the actually tri-linear part of the scene is forced because if they pre-computed the mip-maps and loaded them did they use the exact same formula as they do for auto-generation if not the test is invalid. If they used another filter and auto-generation this test is invalid.

 

[jjayb]

To judge IQ you don't need to know what to look for.

If you want to find worst case scenario's then it can be usefull.

Yes, you do. You just can't look at every possible in-game scenario in any limited time. While I'm sure the shortcomings of ATI's technique will come out eventually, the simple fact is that this technique is against ATI's own claims about what can be considered, "acceptable optimizations," which, in ATI's documents, does not include anything that does less work.

Uhmm, wasn't it "anything that does less work at the cost of image quality"? This isn't exactly manually going through benchmarks and replacing shaders or manually inserting Clipping planes on a known benchmark flyby path. This isn't exactly something that is used just for benchmarks either. You sure do have a way of conveniently twisting things around to try to make ATI look bad.

 

[bloodbob]

Uhmm, wasn't it "anything that does less work at the cost of image quality"? This isn't exactly manually going through benchmarks and replacing shaders or manually inserting Clipping planes on a known benchmark flyby path. This isn't exactly something that is used just for benchmarks either. You sure do have a way of conveniently twisting things around to try to make ATI look bad.

Umm I don't see how Nvidia cheating has to do with this. ATI have done this just because Nvidia have cheated in the past doesn't make it alright for ATI to do the same.

Well nvidia killed 14 children so when the ATI ceo's get sent to court for killing 10 children they shouldn't be found guilty. Does this make sense NO.

 

[Graphics_Krazy]

Nvidia have cheated in the past

NVIDIA are cheating in the present too.
Their filtering is still much lower AND it has no intelligence to default to full trilinear. What you see is their best.

ATI has indicated they are using algorithms to evaluate differences in mip levels AND then reducing from Trilinear Filtering. This IMHO is a revolution in filtering. I hope that NVIDIA's Operation Desperation tactics don't alter ATI's course and work continues on this.

 

[tahrikmili]

I really strained my eyes to figure out that the right side was traditional trilinear, but only after REALLY straining my eyes. Then I went ahead and looked at the posts and found out I was right.

I have to say, if this was not a static shot but a video capture, I would never ever have been able to understand the difference. There is really really negligable difference as far as I can tell..

Yalaz Ozkanli