Rechecking Beyond3D pixel counters

It makes no sense. right. But I'm just saying what I see.
HS 4xAA makes picture look worse than most noAA game do.

No PS3 Hate. Just this technology seems strange to me. Usually AA means AA. And 4x means no jugged edges from playable distance.
if you thought like that because of my SF words... well. SFIV 360 looks better on 360. not a secret for anyone. I don't know why I should not be objective. Still good game on any system.
 
if you thought like that because of my SF words... well. SFIV 360 looks better on 360. not a secret for anyone. I don't know why I should not be objective. Still good game on any system.
Excuse me, haven't you read this?
eloyc said:
"some kind of "PS3 hate" (because of sentences such as "the same strange (I can even say 'terrible') AA I can see in Heavy Rain and GoWIII E3 2009 Demo", and the overall spirit of your words)"
So no, I'm not begging you to be "non objective", there's no need to mix things up by bringing your SF statement here (with which I agree, because it's not a secret, indeed, but that's NOT the point).
 
365 can't you get that MSAA is going "look" better or worse based on things OTHER than the MSAA or platform.
 
365, Humus offers a cool example here of a correct MSAA + HDR implementation. There's even a demo you can play around with.
 
Last edited by a moderator:
yes. I recognized a lot today.
there are "AAs". Sometimes even so strong AA level as 4x can spoil image even worse than absence of AA does.

AA will always give smoother edges and a more 'calm' image unless artifacts appears around edges or it is almost absent due to rendering methods/HW limitations.
 
I am late to the party but as far as I can remember we could get badly anti-aliased edges for at least 3 reasons:

1) Tone mapping is performed after the MSAA resolve pass, which is incorrect because we used a non linear tone mapping operator. But it was way faster so we did it that way :)

2) The shadow term is computed in a deferred pass using a non MSAAed depth buffer, which doesn't exactly match the sample locations in the MSAA4X color pass and can generate false bright or dark areas around certain silhouettes

3) The MSAA resolve pass is done in LogLUV space via texture units, which don't understand that color space and simply perform a linear filtering. Again, we knew this was not correct but we had to do it in order to make it run at proper speed :)

Just have a look at this pic:

heavenly-sword-20070516075649380.jpg


You can see problem n.2 close to the right side of the monster head, where his shoulder/neck silhouette meets that fancy shell (you can see alternating white and black dots if you carefully look at it)

Also you can see how the left half of the image is less anti-aliased due to higher contrast (this is caused by problem n.1) while the right side looks better due to the lower contrast between background and foreground geometry.

Despite these issues I am very happy about the result, we could probably do it much better now, but this tech is 4 years old and it was pretty much unmatched at that time, now even my grandma does fancy color spaces and deferred shadowing :)

Marco
 
Ye gods, look at his teat-sized nipples... Who exactly approved this character design? :LOL:
 
I am late to the party but as far as I can remember we could get badly anti-aliased edges for at least 3 reasons:

1) Tone mapping is performed after the MSAA resolve pass, which is incorrect because we used a non linear tone mapping operator. But it was way faster so we did it that way :)

2) The shadow term is computed in a deferred pass using a non MSAAed depth buffer, which doesn't exactly match the sample locations in the MSAA4X color pass and can generate false bright or dark areas around certain silhouettes

3) The MSAA resolve pass is done in LogLUV space via texture units, which don't understand that color space and simply perform a linear filtering. Again, we knew this was not correct but we had to do it in order to make it run at proper speed :)

Just have a look at this pic:

heavenly-sword-20070516075649380.jpg


You can see problem n.2 close to the right side of the monster head, where his shoulder/neck silhouette meets that fancy shell (you can see alternating white and black dots if you carefully look at it)

Also you can see how the left half of the image is less anti-aliased due to higher contrast (this is caused by problem n.1) while the right side looks better due to the lower contrast between background and foreground geometry.

Despite these issues I am very happy about the result, we could probably do it much better now, but this tech is 4 years old and it was pretty much unmatched at that time, now even my grandma does fancy color spaces and deferred shadowing :)

Marco

Would you go back to work on ps3 games if you have a chance? as working at Intel now does not seem to be an ideal place.
 
Back
Top