[djskribbles]

RB for PS3 has borders but can only be seen if you have a display that has 1:1 pixel mapping.

[Mintmaster]

Originally Posted by warb

Isn't this some kind of AA?

Hmm, some edges do look AA'd and unexplainable by bloom/haze/texturing. I don't know what going on there.

[Cheezdoodles]

Originally Posted by Mintmaster

Hmm, some edges do look AA'd and unexplainable by bloom/haze/texturing. I don't know what going on there.

Same that was going on in GoW perhaps? GoW didnt have any AA (because of UE and tiling or whatever the reason was) but it did have some edge-blur thingy that made it look smoother than no AA.

If my memory is correct

[AlStrong]

Originally Posted by Ostepop

Same that was going on in GoW perhaps? GoW didnt have any AA (because of UE and tiling or whatever the reason was) but it did have some edge-blur thingy that made it look smoother than no AA.

If my memory is correct

Those screenshots are from Gears of War.

[Quaz51]

Originally Posted by AlStrong

Does Commando 3 always render in 1080p or only when you select 1080p in the PS3 settings?

Commando 3 is 1280x720 4xAA for 720p, like X360 version but with really better framerate (very stable on PS3 version and very instable on X360 version)

PS3 720p = 1280x720 AAx4 60fps
PS3 1080p = 1920x1080 AAx2 60/30fps
X360 720p/1080p = 1280x720 AAx4 60/30fps

rare situation

EDIT: it's the studio who make futur SF HD and 1942

[AlStrong]

Thanks for the clarification.

[Cheezdoodles]

Originally Posted by AlStrong

Those screenshots are from Gears of War.

Thought they where GoW 2... Either way, its some edge blur thingy not MSAA as shifty suggested if my memory is correct.

[Shifty Geezer]

Originally Posted by Ostepop

Thought they where GoW 2... Either way, its some edge blur thingy not MSAA as shifty suggested if my memory is correct.

Slight correction - I didn't suggest any explanation! Other than it's not 2xMSAA. It's just something...weird. Not a conventional blur kernal from the looks of it. Maybe a diagonal 2 step blur on the edges? But it seems localised to the stairs. I don't know what technique they were using!

[Quaz51]

it's like an ATI MSAA 2x patern but not visible on all edge and all screenshot

[AlStrong]

http://download.nvidia.com/developer...ed_Shading.pdf

Slide 31 has a suggestion to achieve AA'd edges.

• “Intelligent Blur” : Only filter object edges
• Based on depths and normals of neighboring pixels
• Set “barrier” high, to avoid interior blurring
• Full-screen shader, but cheaper than SSAA

hm....

http://www.freepatentsonline.com/EP1519317.html

The apparatus includes an anti-aliasing system for restoring edge information for an x-direction and a y-direction in screen coordinates from an image after drawing. The apparatus determines a processing content necessary for the anti-aliasing from the obtained edge information and scans either the information of a z-buffer and the information of a normal vector at each pixel obtained at the time of drawing, or both information, or by using the information of normal vectors restored from the information of the z-buffer. Accordingly, the anti-aliasing can be applied to each pixel.

http://jgt.akpeters.com/papers/Rokita05/

Our new idea is to use the Z-buffer directly for extracting visible-edge information. The solution presented here can be summarized as follows: 1) select object edges by applying the Z-buffer spatial convolution with Laplacian; 2) filter out aliasing artifacts by applying low-pass spatial convolution filtering to selected pixels. In this approach, the same circuit architecture can be used for the selection and antialiasing of selected pixels.

This last one I recognize... I think I found it before and posted it here on this forum too!

Edit: Oh rofl.... it was still about Gears of War then too http://forum.beyond3d.com/showpost.p...84&postcount=8

[betan]

Originally Posted by AlStrong

http://download.nvidia.com/developer...ed_Shading.pdf

Slide 31 has a suggestion to achieve AA'd edges.

hm....

http://www.freepatentsonline.com/EP1519317.html
http://jgt.akpeters.com/papers/Rokita05/

This last one I recognize... I think I found it before and posted it here on this forum too!

Edit: Oh rofl.... it was still about Gears of War then too http://forum.beyond3d.com/showpost.p...84&postcount=8

Why would any of those methods look like 2xMSAA instead of 2x2 edge blur ?

[AlStrong]

You probably missed it in the second quote (at the end), but just a guess...

The important bit was how they can select edges. After that you're free to choose a method for AA.

[betan]

Originally Posted by AlStrong

You probably missed it in the second quote (at the end), but just a guess...

Convolution filter on pixels (as opposed to samples) can not look like MSAA,
MSAA is with 0.5, 0,5 pixel shift not 1,1.

And even if it looks like it (but does not), why would anyone use 2 pixel convolution, it's asymmetric and biased against edge direction.

[Shifty Geezer]

Yeah, do you have any example pics AlStrong? Your typical edge detect doesn't create a useful mask that represents overlapping samples, or 'the space between steps'. I don't know of any edge detect method that'll take a signal like this :

---------------- 
                 ----------------
                                 ----------------

...and generate an edge pattern marked out in * like this...

 ----------------********
                 ----------------********
                                 ----------------

If that were readily possible, a blur applied to a geometry determined threshold ought to be pretty effective at providing 2xAA equivalence.

[AlStrong]

The pictures in the document are pretty bad, but yeah, it doesn't look like anything similar to the Gears results.

But after they've detected the edges, could they not have an AA'd render target?

(side note) Have you checked out slides 24 to 30 in the nvidia document? A bit hard for me to understand without the notes

[betan]

Originally Posted by Shifty Geezer

Yeah, do you have any example pics AlStrong? Your typical edge detect doesn't create a useful mask that represents overlapping samples, or 'the space between steps'. I don't know of any edge detect method that'll take a signal like this :

---------------- 
                 ----------------
                                 ----------------

...and generate an edge pattern marked out in * like this...

 ----------------********
                 ----------------********
                                 ----------------

If that were readily possible, a blur applied to a geometry determined threshold ought to be pretty effective at providing 2xAA equivalence.

I'm not sure that's the case for most trivial 3x3 edge detect filters.
I expect them to pretty accurately find all the boundary pixels of both polygons making the edge.
Why do you have discontinuity there?
It seems pretty nice way to generate a mask for blur to me. Of course edge quality won't be any better than full screen blur.

[2real4tv]

This looks ingame for GeOW 2:
http://media.teamxbox.com/games/ss/1...1215358387.jpg

[djskribbles]

yup, looks like it.
also looks the same as GeoW1 (720p, no AA) if i had to take a guess.

[mech]

Originally Posted by AlStrong

On a different note... I just noticed that Rock Band has black borders just like GH2 (360). I'm not sure if it's a lower rendering resolution or if they're squeezing 720p down into a smaller window ala Army of Two on PS3 though.

You can change this in the calibration menu to make it full screen.

[user542745831]

hello,

in which resolution is the orange box rendered in on the xbox 360 and ps3?

is aa applied? if yes, which type?

at how much fps is it running?

couldn't find this info, yet.

[djskribbles]

i think both are 720p. the 360 version has no AA, and i think the ps3 version has quincunx. both are 30fps, but the 360 version has a more stable framerate.

[thambos]

Originally Posted by user542745831

hello,

in which resolution is the orange box rendered in on the xbox 360 and ps3?

is aa applied? if yes, which type?

at how much fps is it running?

couldn't find this info, yet.

http://www.eurogamer.net/article.php?article_id=89373

[Betanumerical]

Originally Posted by djskribbles

i think both are 720p. the 360 version has no AA, and i think the ps3 version has quincunx. both are 30fps, but the 360 version has a more stable framerate.

That is very odd to say the least.

[TheBlackLodge]

Eurogamer had some comparison shots between all 3 versions and the PS3 version was sadly blighted by the 'vaseline filter' effect.........

[djskribbles]

why? because the ps3 version has some form of AA and the 360 version doesn't, and that the opposite is usually true? i thought it was weird too, but i guess quincunx comes at little cost for RSX... or am i wrong?

if you look at the eurogamer comparison, aliasing is definitely worse on 360, but clearly the ps3 version is a little blurrier. so you would think the obvious answer would be quincunx.

Originally Posted by TheBlackLodge

Eurogamer had some comparison shots between all 3 versions and the PS3 version was sadly blighted by the 'vaseline filter' effect.........

i think people said that only EP2 had this "vaseline filter" effect, and it was excessive motion blur.