Alternative AA methods and their comparison with traditional MSAA*

But You can be right that they havent got skills for that..

Don't be silly. It's just a shader! Just because they haven't implemented it in the full game doesn't make them incompetent - as you know, it has to be integrated properly, not just some silly hack. You might think they have other pressing matters to attend to as well.

edit: Their temporal reprojection + post-AA is already more involved...
 
How exactly? A solid example would be nice.

- Faster comparisons in different games, so also different tech/art style, I know that You can record noAA version and then process it with FXAA/MLAA, but it isnt too comfortable.
- Realtime preview on AA and it cost on GPU
- Higher control group, because everyone can use it
 
found out that the FXAA injection method works on VLC player, we can how use it to process images and videos if anyone is interested to know. Could be faster to help out 3d rendering by skipping the AA step. Interesting enough that I took the new Gamescon U3 gameplay footage, the highest quality one that I can find on the net and tried it. It helps a little bit on the distance objects which MLAA failed to pick up bit the overall IQ definitely take a hit (blurred a bit).
 
Well the FXAA beta10 Injection DLL hack works really well in Deadspace2. Clean up lots of very visible aliasing that the ingame AA mode did nothing for!

This is the best thing for PC Gamers in quite some time. Of course I'm really looking foward to developers running with this... in their new titles.
 
Could someone who has the latest (beta10?) injection mod please upload it somewhere like mediafire? I have serious problems using the guys regular download site. TIA

Not sure why you'd want it to work in TES:IV. With MSAA + AAA/TAA in the driver panel it already looks sweet, IQ wise anyway.
 
Could someone who has the latest (beta10?) injection mod please upload it somewhere like mediafire? I have serious problems using the guys regular download site. TIA

Not sure why you'd want it to work in TES:IV. With MSAA + AAA/TAA in the driver panel it already looks sweet, IQ wise anyway.
More concerned about performance than I am about IQ, since my PC can't even handle 4xMSAA without speed issues. FXAA is much better in this regard.

Beta 10: http://dl.dropbox.com/u/12527604/stuff/injectFXAA-beta10.zip
 
Has there been single demo on subpixel capable AA techniques like SMAA or SRAA?
It would be very nice to see different postprocess AAs on single demo and check the quality differences.
 
It would be very nice to see different postprocess AAs on single demo and check the quality differences.
My opinion after some fooling around with them in a few scenes:
- GBAA/DEAA are generally the fastest and look pretty good if you have fairly big mesh triangles. They start to lose AA and introduce distracting noise in cases with small or sliver-y geometry.
- MLAA does a good job on long edges and is fairly cheap these days as well.
- FXAA does a slightly less good job on long edges but is very fast. The subpixel contrast thing is mostly a gimmick as it only really does anything to frequencies that happen to be around the screen sampling limit. Higher frequencies obviously still flicker in and out with nothing FXAA (or any screen-space method) can do about it. The low pass filter over these areas actually makes it look worse in a lot of cases too as we've seen with DoF over aliased regions in the past.

But the real kicker... they all look like crap compared to even 4x MSAA. So yeah, if you can't afford 4x MSAA and need something quicker then they make sense, but I doubt MSAA is going anywhere.
 
But the real kicker... they all look like crap compared to even 4x MSAA. So yeah, if you can't afford 4x MSAA and need something quicker then they make sense, but I doubt MSAA is going anywhere.
...unless you're talking about foliage. Could FXAA be seen as a super cheap replacement to TAA/AAA? Used in combination with some MSAA it should be pretty nice overall.
 
...unless you're talking about foliage. Could FXAA be seen as a super cheap replacement to TAA/AAA? Used in combination with some MSAA it should be pretty nice overall.

FXAA can never replace TAA. That is actually a case where it fails (subpixel stuff like fences), though not as bad as some other PP AA methods.
 
...unless you're talking about foliage. Could FXAA be seen as a super cheap replacement to TAA/AAA? Used in combination with some MSAA it should be pretty nice overall.
For foliage you want blending, plain and simple. See Marco's Adaptive Transparency paper :) Alpha to coverage with better patterns and more samples is an option too.

And as homerdog mentions, foliage is actually a really bad case for the screen-space methods since it has tons of subpixel frequency stuff. Again, any large continuous edges are just... not hard.
 
Well the FXAA beta10 Injection DLL hack works really well in Deadspace2. Clean up lots of very visible aliasing that the ingame AA mode did nothing for!
There are a couple of ways (for example) to get SSAA working for that if you have a NV card. There were some caveats though, with the lighting and shadows. I'm not sure if anyone got it working perfectly. The original game doesn't have any problems with forced NV SGSSAA. I tried MLAA on it too but it was just laughable by comparison.
 
Shame the injected FXAA doesn't work all the time (though it's not surprising). I'm trying to get it to work with Freelancer, since it has issues with regular AA, but it's refusing to load.
 
I gave Dead Space 2 a go with the FXAA injection and yeah it is nice. It's great to have at least this option.
 
Back
Top