Image Quality and Framebuffer Speculations for WIP/alpha/beta/E3 games *Read the first post*

Drive club looks to be without any multisampling, after watching the footage.

I Downloaded the 1.5 gb file, media compression is still getting in the way. These are the best i can do. (media sharpen filter was set to max.

OEab.png
)
QEab.png

TEab.png

UEab.png
 
Is it possible they are using "vertical interlace" a system such as "horizontal interlace" on KZ shadow fall?

No. It's definitely the encoding solution, probably a 1080i encoding creating artefacts of a 540p image. In fact KZSF MP is 960x1080p when in moderate/fast motion when motion vectors can't properly create a native 1920x1080 image.

I have already seen such 1080i / 540p artefacts with a Forza 5 and also Watchdogs video before.
 
I Downloaded the 1.5 gb file, media compression is still getting in the way. These are the best i can do. (media sharpen filter was set to max.

I get conflicting results and I am not certain about thoses results:

The indoor image looks strongly 1080p native (only horizontally verified but probably vertically too) but the outdoor similar pics possibly look ~800p horizontally and vertically verified, but beware, it could be between 720p and 900p and I am much less confident than with the indoor pic. Don't quote me! we need much more screenshots!

media sharpen filter was set to max.
Please don't do that! I am prettty confident about the indoor native 1080p horizontal resolution, could you post more outdoor images (and indoor with long horizontally slanted edges) without the sharpen filter (if possible of course)?
 
The indoor image looks strongly 1080p native (only horizontally verified but probably vertically too) but the outdoor similar pics possibly look ~800p horizontally and vertically verified, but beware, it could be between 720p and 900p and I am much less confident than with the indoor pic. Don't quote me! we need much more screenshots!

Different resolutions between indoor and outdoor??! Well that'd be odd.

I would have just assumed it'd be best-foot-forward PC version with Xbox controller and no need to analyze at all.
 
Unfortunately those are not helping much, but thank you for your effort! Also don't post 3 times the ~same screengrab that you already posted before^^.

Ideally only big views of Paris with roofs, plenty of contrasted straight lines (but a bit slanted). When he is on the roofs, 3 different screengrabs should be enough.
 

I have just read those slides which really worried me in the future of AA in Ubisoft future games.

Notably this flipquad, FQ blur algorithm (their own words "0.5 blur") that seems to be intergrated in this new HRAA. They even compare it with the infamous vaseline Quincunx.

So FQ blurs so much the textures that they must use a sharpen effect afterwards:

njbb.png


They even admit themselves that the blurring effect might by a problem for the art direction (meaning it might blur more than even FXAA!) :

mjbb.png


They are really going to ship FC4 with an integrated blur + sharpen effect? Do they know that a sharpen effect in fact create a false "sharp" effect by destroying the real information, easily visible on this sharpened texture?

The 1ms version doesn't use SMAA anymore (contrary to AC4 and Watch_dogs), morphological AA is too costly (maybe to much memory costy also, this HRAA must adapt to different hardwares...)

This HRAA, to me, looks like the new FXAA. A temporaly stable FXAA and it worries me.
 
I hope I'm not derailing the thread, but what exactly is a framebuffer? What's a "buffer"?

A framebuffer would be a portion of memory that contains the data that would be pushed to your screen. It would contain the data for each pixel.

A buffer is a generic term for a section of memory that is used to hold data while it is being transformed or moved.
 
They are really going to ship FC4 with an integrated blur + sharpen effect? Do they know that a sharpen effect in fact create a false "sharp" effect by destroying the real information, easily visible on this sharpened texture?

The 1ms version doesn't use SMAA anymore (contrary to AC4 and Watch_dogs), morphological AA is too costly (maybe to much memory costy also, this HRAA must adapt to different hardwares...)

This HRAA, to me, looks like the new FXAA. A temporaly stable FXAA and it worries me.
I'm not quite sure why you mention FXAA, due to additional blurring like in AMD tent filters and TXAA?

They either use a lot coverage samples and/or AEAA(GBAA) to get more stability, both actually give sub-pixel information.

More complex reconstruction is the way to get a lot more quality from performance used and is a right way forward, even if MSAA samples are used. (Like Order 1886 seems to do..)
UE4, temporal AA.
 
Last edited by a moderator:
Flipquad is much better than quincunx. It provides similar improvements over quincunx to edge quality (edges passing pixel) than rotated grid AA provides versus ordered grid (roughly double sample quality for near vertical/horizontal edges). It is also more stable and it hits thinner geometry better than quincunx or 2xMSAA (post AA is much worse in these than any real multisampling). So it has some good qualities and the cost is equal to 2xMSAA. However it blurs slightly...
 
I'm not quite sure why you mention FXAA, due to additional blurring like in AMD tent filters and TXAA?

They either use a lot coverage samples and/or AEAA(GBAA) to get more stability, both actually give sub-pixel information.

More complex reconstruction is the way to get a lot more quality from performance used and is a right way forward, even if MSAA samples are used. (Like Order 1886 seems to do..)
UE4, temporal AA.

The rest of HRAA (Hybrid RAA...) do look great (on paper). But not the FlipQuad (modern Quincunx vaseline filter) stuff.

Blurring a texture and use a sharpen filter after is awful, terrible. Why I am the only one to be shocked by the procedure, here at B3D? Even if you'll get textures as sharp and ~identical as the original (on your couch at 5 meters from your TV), you'll loose ~75% of the initial high frequency details and inevitably there'll be nasty artefacts.

They even admit in the slides that art direction might complain a bit so the image quality must be altered somehow. Also in the slides they never really showed the result of HRAA with real full 1080p detailed images, neither compared it to others solution like for instance SMAA slides which did it exhaustively.

EDIT: Just saw Sebbbi post which confirmed me my worries. Using a blur filter is a cheap way to get rid of jaggies. But destroying 75% of high frequancy details in order to get rid of 1% of image's artifacts (jaggies) is a bit extreme.
 
Back
Top