Xbox Upscaler... Any Details and Information?

[djskribbles]

Unless dark10x altered the images, but I have no reason to believe so. I took the links straight from dark10x's post.

Personally, I think that even Lanczos is a bit too sharp. For reasonably fast scalers, I prefer Spline slightly more than Lanczos. On my HTPC, I use Jinc to upscale videos (with madVR) which is pretty GPU intensive, but I use Spline on my slower PCs.

Here's the 720p image scaled using Spline36 instead:

Spoiler

Looks pretty much the same as Lanczos unless zoomed in, but it's slightly softer.

 

[RudeCurve]

When comparing between those different images I'm not only looking at the outlines of objects but also the clarity of small details throughout said objects. I think if the goal is to scale without losing sharpness while not increasing sharpness then the Spline36 method looks the closest to the native 720p.

I think it would be nice if MS adds a scaler sharpness adjustment option in the systems settings in the future.

 

[PixResearch]

I just pulled those into Photoshop CC as it's all I've got to hand. The xbox result looks near identical to photoshop's default upscale which I *think* is a bicubic…


As an aside:
the lanczos image provided is waaay softer than it should be. Whoever did that upscale used a 4 lobe version of the lanczos filter when they should have used a 2 lobe version. Unfortunately lots of textbooks make this mistake so it's made its way into image and video scalers. Wider filters make perfect sense in audio, etc where ideally they should be infinitely wide but in images you are just sampling pixels further away from and messing up the image quality.

 

[MBDF]

Of those above images, I would prefer spline as well.

This may be off topic but any news or information on the PS4's scaler?

 

[Steve Dave Part Deux]

Yeah, why would anyone question the integrity of GAF as a reliable source of Xbox news? I have a hard time imagining that they fudged the scaler given the hardware was designed to exploit it's use extensively. You're approaching this whole argument from the perspective that Sony's engineering acumen is far superior to Microsoft's. Last I checked, Sony's last major engineering effort involved a multi-billion dollar investment being thrown into the Pacific Ocean.

 

[djskribbles]

I just pulled those into Photoshop CC as it's all I've got to hand. The xbox result looks near identical to photoshop's default upscale which I *think* is a bicubic…


As an aside:
the lanczos image provided is waaay softer than it should be. Whoever did that upscale used a 4 lobe version of the lanczos filter when they should have used a 2 lobe version. Unfortunately lots of textbooks make this mistake so it's made its way into image and video scalers. Wider filters make perfect sense in audio, etc where ideally they should be infinitely wide but in images you are just sampling pixels further away from and messing up the image quality.

I resized them, and I stated in my post that I used Lanczos4. What do you mean it's way softer than it should be? In reference to what?
FWIW, I used avisynth filters to resize the image in AvsPmod. Regardless, I greatly prefer the Lanczos4 and Spline36 images to the X1 scaler image. The X1 scaler introduces way too many artifacts (aliasing, ringing, rough edges).

I also doubt that the X1 just uses bicubic. I've never seen bicubic introduce so many artifacts. MS says that their scaler is much more advanced than the X360. I'm guessing they use some custom scaler that has very aggressive sharpening.

 

[RudeCurve]

This is a straight bicubic in PaintShop Pro X6...set to 50 which is midway between soft/sharp (no sharpening). This looks exactly like the native 720p source.

 

[djskribbles]

Why did you scale it to 1600x900? Makes it difficult to compare.

 

[RudeCurve]

Hmm weird I guess I overlooked the settings I try again. Actually my image is 1080 but I guess the Tinypic.com site scaled it down.

 

[djskribbles]

Just realized that imgur converted my above images to jpg. Gotta re-upload them.

edit: fixed with proper pngs.

 

[PixResearch]

I resized them, and I stated in my post that I used Lanczos4. What do you mean it's way softer than it should be? In reference to what?

I mean Lanczos4 isn't the scaler someone who was trying to build the highest quality image scaler should be using. A narrower Lanczos filter (i.e. Lanczos2 filter would be both cheaper to implement in hardware and would look better).

The "reference" is that the Lanczos image you've shown is softer than the other two images. Lanczos is typically at the sharp end of scalers whereas you've shown it as the softest image. A product of using a Lanczos4.


Note: It's probably 6 years since I looked at scalers properly but the engineers at Microsoft/AMD will know this stuff inside out. If they had the silicon budget they'll have gone for either a lanczos2 or Catmull-Rom (they look near identical). You can do better with image registration and super sampling but they won't have gone down that route on a console. In reality I suspect most people can't see the difference while things are moving and the quality of the display would have a bigger effect than pixel peeping.

 

[brunogm]

Upscaler testing framework

The EXQUIRES (EXtensible QUantitative Image RESampling) Test Suite
http://exquires.ca

Maybe can be used to discover what is the scaler on X1, by using correlation tools?

 

[Cyan]

So these shots were posted at GAF of KI running at native 720p output and 1080p output through the X1 scaler:
http://www.neogaf.com/forum/showthread.php?p=91025042#post91025042

I took the 720p native image and scaled it to 1080p using Lanczos4.

720p native output:

Spoiler

1080p output (X1 scaler):

Spoiler

Resized using Lanczos4:

Spoiler

Lanczos4 is a pretty standard scaling algorithm that does a pretty good of upscaling images. It is known to introduce some artifacts from minor sharpening. But you can see the Xbox One scaler introduces a lot of rough edges and ringing from over-sharpening when compared to Lanczos4. Lanczos4 does a much better job here IMO. I've said this before on these boards, but again, I'm not a fan of the X1's scaler at all.

I didn't notice much sharpening when playing KI.

It is more obvious in CoD: Ghosts and AC IV -played each game like 10-20 minutes. I thought there was some sharpening before reading anything about NeoGAF and stuff.

When playing those games I noticed that and I gotta say that it is a high quality upscaling, there is no reason to believe otherwise.

 

[mikiex]

The EXQUIRES (EXtensible QUantitative Image RESampling) Test Suite
http://exquires.ca

Maybe can be used to discover what is the scaler on X1, by using correlation tools?

Unless its programmable as it was on the XB360, where the developer can choose and tweak the method.
Not that I know if any developer bothered changing it from the defaults.

 

[Pixel]

Still curious how the [strike]40nm[/strike] 28nm 7xxx Pitcairn upscales in comparison to Xo's custom scaler. Might assume PS4 upscales in the same manner as the Pitcairn chips.

edit:Thank for the fix, somehow i wrongly ended up thinking half the south isle lineup was fabbed at 40nm including the 7850

 

[Deleted member 13524]

Still curious how the 40nm 7xxx Pitcairn upscales in comparison to Xo's custom scaler. Might assume PS4 upscales in the same manner as the Pitcairn chips.

There's no 40nm Pitcairn. All Southern-Islands GPUs are 28nm.

 

[Pixel]

Call of Duty Ghost:

Prepatch: Both PS4 & Xbox versions are 720p (both upscaled)
Post-patch: PS4 rendered at 1080p


XBox One Xbox 720p upscaled to 1080p

Spoiler

PS4 1080p reference:

Spoiler

Eurogamer PS4 unpatched 720p

Spoiler

Eurogamer PS4 patched 1080p

Spoiler

Image 1 and 3 can be used for scaler comparisons by comparing the dog and the tank

 

[djskribbles]

The X1's scaler seems to be able to apply different levels of sharpness because they removed it from the retail version of BF4 (unless they switched to software scaling like Crytek did for Ryse). BF4 PS4 doesn't have a 'vaseline smear', it just looked soft compared to the PC version (which wasn't as soft because it had better AA and was running at a higher resolution) and the aggressively sharpened Xbox One version. If you look at the recent BF4 DF comparison, the PS4 version looks a bit sharper than the X1 version now that the sharpening has been removed.

IMO, CoD X1 doesn't look like it's applying as much sharpening as say, BF4, which is why the PS4 and X1 version looked similar before the PS4 version was patched to native 1080p. So maybe the scalers are similar, as long as they don't apply that nasty sharpening.

edit: you can directly compare the pre-patched PS4 version (720p) to the X1 version here:
http://www.thesixthaxis.com/2013/11/12/call-of-duty-ghosts-xbox-one-comparison-video/

X1:

Spoiler

PS4 pre-patch:

Spoiler

X1:

Spoiler

PS4 pre-patch:

Spoiler

Looks like a bit of sharpening on the X1 version IMO, but they look fairly similar.

 

[Pixel]

Thanks and thanks for that analysis.

 

[fehu]

ModEdit: context
http://www.theverge.com/2014/2/12/5...-the-xbox-one-controller-for-less-twitchy-aim

Respawn Entertainment also says it's working with Microsoft on a new version of the console's upscaler

How much hardware is in the hardware scaler if you can fix with an update?