Alternative AA methods and their comparison with traditional MSAA*
- Thread starter mitran
- Start date
Comparison shots 2 suggest that Pandemic uses edge detect on PS3 with a much higher threshold, so that only clearly contrasted lines are dealt with. The Intel method considers almost anything an edge and blurs the whole picture. Look at the huge difference between the boxes. Because the boxes have dark-colored edges on the outside almost nothing is done with them. On the inside there's a lot more contrast and so more processing it seems. It also seems to be an almost exclusively horizontal effect.
Also when rendering, the developers have access to more than just the luminance or RGB to edge-detect. More selectivity should be a big plus, elliminating unwanted blur or artefacts with problem situations.
Silent_Buddha
Legend
Good AA implementation is always more noticeable in motion than in static shots. It's one reason why AA was so easily discounted in the past. The whole "well you have to magnify the image to notice aliasing on such and such edge" edge for example.
While that same edge in motion needed no magnification to notice the artifacts as they crawled/sparkled/etc... And the issue was even worse when you had multiple parallel lines in close proximity to each other and then you moved the view... /shudder.
Regards,
SB
I saw Christer Ericson's nick appearing twice on B3D online members list for the past few days. Checked his Twitter:
http://twitter.com/ChristerEricson
No further words from Mr. Ericson besides his quick comments ? The Saboteur's implementation looks different from the Intel one (Just edge detect + blur ?)
Most likely theyre also use the depth values/(or normals perhaps) as well to see if something is truly an edge.
I've gotta admit this technique does look great, but will have to see it in other games first before a clearer picture can be formed, with saboteur the images have way too much DOF blur to be able to form a true opinion.
Ive got a feeling this is gonna be the next 'in thing' for games to have, SSAO was the previous
Eh? Can't expect any details from a tweet of 120 chars! This the one you're refering to?
@kamidphish edge filter plus blur/blend
3:32 PM Dec 12th from TweetDeckWhich was a response to:
http://twitter.com/kamidphish
@ChristerEricson We did that too. Can we be part of the cool crew? about 19 hours ago from TweetDeck in reply to ChristerEricson
What is this magical Saboteur "AA" method? 1:22 PM Dec 12th from TweetDeck
He already clarified with DigitalFoundry in slightly more details, but I was hoping for more:
From his twitter, it sounds like classifying it as Intel's MLAA is off (PeterT mentioned that too).
If his method produces equally impressive result, then yes, they may be in the same club. Does it
The magic sauce is in *how* they detect edge and perform blur; not whether they detect edge and perform blur or not.
From his twitter, it sounds like classifying it as Intel's MLAA is off (PeterT mentioned that too).
makattack said:
If his method produces equally impressive result, then yes, they may be in the same club. Does it
The magic sauce is in *how* they detect edge and perform blur; not whether they detect edge and perform blur or not.
If Fragment-AA maps to the SPUs maybe Sony should go to Matrox, which is still an independent company, and license the patents.
user542745831
Veteran
I'd be interested in seeing the effect on the explosions in Infamous. Normally when we do the 1/4 size particles on PS3 games, we add some kind of post process on them to try and smooth out the rough edges a bit. That takes a bit of time to do though. So I'm wondering if a luminance based AA method would have a smoothing effect on the rough transparency edges, and then maybe the post process we do on the transparency buffer could be totally discarded. I've done a Z based edge blur in the past on PS3, but that of course doesn't do anything on the transparencies since they don't write to Z, hence I'm now curious if the luminance method would take care of it. Infamous would be a good test case for this since they don't do any post process on their transparencies and hence they look really rough, so before and after shots of Infamous with luminance AA applied could be revealing!
Matrox's FAA isn't a post process step. Edge fragments append to a buffer as they are rendered. Without this information the technique doesn't work. Plus, it had multiple flaws so I doubt anyone would want to just license it.
It seems like a lot of effort for questionable gain. The GPU would need to know if the pixel shader is working on an edge and export coverage, color, and Z info. Even if it could work it would likely be very slow. It definitely gets away from the idea of using the SPU as this puts more work on the GPU just to have the SPU do a bit of post processing.
Plus, there are aspects of FAA that I doubt the PS3 GPU could implement.
Well FSAA is not a technique at all, it just implies that some form of AA is being applied to all the pixels on the screen. Full Scene Anti-Aliasing
http://www.playstationuniversity.com/improvements-to-gowiii-since-demo-explained-2774/
“The E3 content is over a year old,” Feldman writes on the God of War forums. ”The team has learned a lot since then, so the final game should reflect that effort. Off the top of my head features added since E3 include: anti-aliasing on the CPU – looks much better then the 2x at E3"
could it be the same trick?
“The E3 content is over a year old,” Feldman writes on the God of War forums. ”The team has learned a lot since then, so the final game should reflect that effort. Off the top of my head features added since E3 include: anti-aliasing on the CPU – looks much better then the 2x at E3"
could it be the same trick?
I doubt that it is the same, but it could be something similar.
We will know when we have our dirty hands on the game.
Similar threads
