AMD FSR antialiasing discussion

Hasn't it been established that with DOF disabled FSR still looks worse than TAAU in KingsHunt? So while you have a point it seems umm pointless now?

 

The developer gets to choose how much sharpening is applied as a part of the FSR package.

 

No it hasn't been established, but glad you are at least recognizing the problem now even though you are trying to hand-waive it away still.

I'm not surprised that you are only looking at FSR's worst case scenario, one that AMD itself even recommends against using unless absolutely necessary.



https://community.amd.com/t5/blogs/amd-fidelityfx-super-resolution-is-here/ba-p/477919

Are to tell me what their recommendation is for performance mode vs all the others?

But thank you for at least acknowledging that the TAAU comparison is invalid because of the breaking DOF.


https://ibb.co/k4DDYXK vs https://ibb.co/0jnnjB9

https://ibb.co/k0Q444C vs https://ibb.co/svWL4tv

 

I see you like sophistics over facts.
Here are a few facts - sharpenning doesn't add any texture or geometry details, another fact is that FSR oversharpens image to a point where it appears sharper than Native and that's not necessary developers intended. Why would developers intend to piss off all those people playing in native resolutions by not adding tons of sharpening in native if it was all good?
Oversharpenning is not a good thing - it adds ringing, posterization, amplifies any moire, shimmering and all kinds of aliasing in motion, breaks down gradients of anti-aliasing and introduces other artifacts, the funny part is that spatial upscaling exaggerates any kinds of temporal artifacts as well, so while it can look good on your static screenshots, it can as well be totally broken during gameplay.
I would not like developers to select the sharpness level for me because there is no way to tune the sharpness down once it's added without degrading image quality even further. Even if they don't add sharpening control knobs, I have my own means to tune up the level of sharpness to whatever level I would consider appropriate via the sweetfx, reshade, freestyle, etc.

It has been, but he will simply ignore that fact as all others.

 

You claim that FSR oversharpens the image, but the developers choose the sharpening level. So the developers do want it to look that way. They can add a sharpening slider if they wanted to.

Problem is once again, you are failing to admit that TAAU testing was bugged and invalid because of that. Hell even Alex from DF admitted it was bugged and he messed up, he just never fixed his articles to show the real difference.

 

The lack of the sharpening slider and control commands in all games says otherwise.

It wasn't invalid at all - https://forum.beyond3d.com/threads/amd-fsr-antialiasing-discussion.62394/page-33#post-2212099

 

You sound like a broken record. Was that your reddit post that was criticised?

 

How does a lack of a slider mean that the developer didn't select the sharpening level? In fact, it proves that is the amount they wanted because they aren't letting you change it.

Look at the images posted above, along with the message from AMD itself saying that performance mode has the worst image quality and should only be used if absolutely necessary.

https://forum.beyond3d.com/posts/2212528/

 

Isn't DLSS also using sharpening? Even more, most people on reddit will tell you to use more sharpening on it, if it looks blurry. Yet no one said you need to sharpen the native image or other upscaling method before comparing it do DLSS.
And also the problem is with only comparing 1080p to 4k upscaling. On a podcast with Frank Azor, the host said that one day before FSR was launched, Nvidia had a conference with some youtubers and told them how hard it is to upscale an image from 1080p to 4k, suggesting somehow that they should insist on FSR performance.
And yes it is a valid comparison but it is not the only one, you should at least try to compare the higher res versions before dismissing everything saying that it is the same as it is at 1080p -> 4k
Or saying that the TAAU artefacts don't matter because FSR will have them too. How can it have these artefacts since TAAU was not even included in the game?
Every time a new version of DLSS was released, we have seen all performance steps being compared with native resolution ( or other forms of upscaling ). Control 1.9 version was considered so great at that time compared with previous versions of DLSS that Alex said it is better than native, it is written in the DLSS topic. Pretty low standards at that time from him and a lot of IQ experts. Good thing they have better standards now.

 

I must sound like a broken record because some people here can't seem to admit that DOF being broken is a major problem in DF's testing. Not my thread, but I agree with the author of it and shocked how many people don't seem to care about integrity in reviews. Do you work at NV with Oleg?

 

And what do you call a overshapened but undersampled image? Do you think the developer of the Terminator game thinks that a native rendered image should not at least as sharp as an undersampled upscaled image?

TAA and TAAU, too.

DLSS is a substitute for TAA or TAAU. FSR is just a lower resolution image upscaled to native. Why should it be possible that this upscaled image could be more sharper than the native rendering?

 

Is FSR with DOF disabled in KingsHunt producing a noticeably worse image than TAAU with DOF disabled? Yes or no?

The rest of it seems to be a crusade to get people (which people?) to admit that one specific DF comparison image was flawed. Which basically everyone including DF has already admitted. So the crusade can wrap up now.

 

Time stamped. Alex explains himself quite well. Worth a listen for anyone that's raging about the DF coverage.

 

Because it is using sharpening to obtain a desired effect? Why is DLSS using sharpening since it is not a lower res image but the result of many frames + AI trained on 16k images? What kind of question is that? Both features are using sharpening because that way it is easier to fool gamers that their upscaled images are similar to native image. Yet DF never sharpened the native picture to make a "fair comparison" like they did now with FSR.

 

You tell me, the bottom of that post you just quoted has comparison images.

Please point out with specifics which areas you think look better or worse and why.

 

Didn't told us if he was briefed by Nvidia to insist on the FSR performance or at least if it was briefed by Nvidia how to test Control 1.9 in order to reach the conclusion that it is better than native.
The story about Nvidia briefing one day before FSR launch is in the full nerd podcast at 8:15 minute.

 

DLSS doesnt use additional sharpening, the network creates a certain sharpness level. In most cases it is not as sharp as the native rendering with TAA:

The DLSS upscaled image is not undersampled unlike simple in engine upscaling.

 

Sorry but that is plain stupid. The fact that the network is the one that adds sharpening ( or creates a certain sharpness level ) means nothing. It is additional sharpening of the image. If DLSS has a sharpening pass, it is does the same thing FSR is doing.
The idea that you can improve the native or other upscaling method and then compare it with the upscaler you dislike, is new. Why not writing some new motion vectors for UE5 TAAU and then compare it with DLSS and prove DLSS is not so great because it has more ghosting than your modified TAAU.

 

Do you think Alex did this video in 1 day?

This is getting more and more ridiculous.

And all that "AMD recommends against usinig Performance" stuff is just hilarious.

Of cause they do, that's a spatial upscaling after all.

Have you tried reading the "A Survey of Temporal Antialiasing Techniques" paper?
I saw people already suggested you reading that.

It's really painful to see how you guys have no basic knowledge of image processing methods, yet accusing Alex of some really stupid stuff you find appropriate.

You have to understand the difference between spatial upscaling and temporal accumulation, otherwise there is no point in talking with you.

 

DLSS does has sharpening feature, and its developer controlled, some games use it some dont, maybe quality mode does not use it, other modes might do. I cant link the twitter right now, here is the screen shot

If FSR is just sharpening to "fake" better than native look, then how is it resolving and smoothing the jagged branch and tree leaves thats present in native, in your comparison image?



The amount of hoops some people willing to jump to "prove" their illogical point is just baffling.