Star Wars GOUT in HD using super resolution algorithm (* unfinished project *) — Page 37

Here's a split screen version of the sample video. Left is SRV12, right is a PointResize upscale of the source.

https://drive.google.com/file/d/0B8_LYKyZDiajVEtVRkRwNXlCYkE/view?usp=sharing

I will try to do a short clip of SRV11 raw during the weekend. 

Very cool!  It will be neat to at least see a small amount of V11 in action.  The scenes from 5:32 to 6:17 would be my choice.  I want to see Vader's forearm in motion while he is holding the guy up.  I think it has more detail in V11.  I would also like to see 2:30 to 3:34.

What else is new?  Are you rendering the rest of the movie?  Thanks :-)

I've been making some backups. The rendering of the full movie is next on the agenda. That will probably start tomorrow. 

Every once in a while I accidentally click the first page of the thread instead of the last, and see the post,

"this will look like shit and I know exactly why."

It's hilarious every time.

Funny!  It got me reading the beginning pages of this and came across this one on page 3.

"Super Resolution only exists in papers of shortsighted academics who have a poor grasp on sampling theory."

At least he somewhat redeemed himself at the bottom of the same page.

thorr said:

Funny!  It got me reading the beginning pages of this and came across this one on page 3.

"Super Resolution only exists in papers of shortsighted academics who have a poor grasp on sampling theory."

At least he somewhat redeemed himself at the bottom of the same page.

 That user at least has a lot of experience in trying to clean and upscale the GOUT, including an SR attempt, I believe. Though, IMO, the first guy redeems himself by having a Pink Floyd avatar as well. :)

RU.08 said:

It certainly isn't SR on its own that brings the source to life - its his Avisynth script that implements SR in his own uniquely designed way.

Exactly. That's what I was saying from the very beginning: the substantial part of SR adds very little to the overall effect. It is the noise canceling and sharpening/deblurring that happens during the process are responsible for most of the improvements.

And, of course, the smart anti-aliasing elevated the earlier results to another level. It is very impressive indeed.

thorr said:

Funny!  It got me reading the beginning pages of this and came across this one on page 3.

"Super Resolution only exists in papers of shortsighted academics who have a poor grasp on sampling theory."

At least he somewhat redeemed himself at the bottom of the same page.

 I stand behind what was stated. Only a very small percentage of what you are considering improvements here are due to SR. What you have is a denoiser that leaves a sheet of slowly moving grain, oversharpening effects (sorry, de-blurring artifacts), and shimmering residual aliasing. What's happening here does nothing to restore the rivets on R2, actually blurs Obi-Wan's hair into a solid mass where individual strands were once visable, turns most objects into oil paintings, and only fakes out one into thinking there is more detail through edge enhancement. 

But, please, by all means, consider me an asshole. I really don't care.

-G

g-force said:

But, please, by all means, consider me an asshole. I really don't care.

-G

 ok. I don't think he was calling you that, but can probably accommodate if you like.

g-force said:

But, please, by all means, consider me an asshole. I really don't care.

He didn't. 

It's a good thing DrDre didn't take such offence at being referred to as a "short sighted academic".

Welcome back g-force! There are certainly artifacts, there's no denying that. However, there have been plenty of comparisons to other upscaling methods that show that there is a substantial amount of detail recovered. Whether the gains outweigh the losses is a matter of opinion. You obviously don't like the outcome, and that is fine. Many members are pleased with the results or feel the methods applied are useful to their own upscaling projects. That's good enough for me.

towne32 is doing a substantial cleanup of the GOUT Star Wars as we speak and should be finished in a few days, so I will be using that as a starting point for the rendering of the whole movie. In the mean time I will try to make a short sample for SRV11 and post a few screenshots for TESB. 

RU.08 said:

g-force said:

I stand behind what was stated. Only a very small percentage of what you are considering improvements here are due to SR. What you have is a denoiser that leaves a sheet of slowly moving grain, oversharpening effects (sorry, de-blurring artifacts), and shimmering residual aliasing. What's happening here does nothing to restore the rivets on R2, actually blurs Obi-Wan's hair into a solid mass where individual strands were once visable, turns most objects into oil paintings, and only fakes out one into thinking there is more detail through edge enhancement.

 What utter bullshit. "Only fakes out one into thinking there is more detail ..." Of course there isn't more detail, but rather the detail that is present across several frames is recovered. The net effect is that each frame contains more detail than it did previously.

SR has quite limited use, obviously, but this is one source that clearly benefits from it.

 Okay, let me ask you this. How much of the improvement is from the temporal combination? How much is from all the other things that the script is doing? Scientists usually publish their methods, I have yet to see what the script is actually doing. I guess I just fundamentally have a problem with applying a bunch of denoising and spatial filters and claiming that it's all SR.

-G

This is a funny debate considering "we" (g-force, RU.08 and me) all say essentially the same thing.

I certainly do not know enough to determine how much of the improvement is due to simply sharpening vs. the SR, or what is a due to the combination, that sharpening alone couldn't achieve.

I do know the last 3-4 versions of the script make the GOUT, in my opinion, look better than other attempts that I have seen. I chuckled at the guy who said "this will look like shit" rather than g-force, because the other guy turned out to be quite wrong. g-force: you make some good points, but I am not sure that the claims that you take issue with are really being made. If it was "all" SR, I think this would have been completed in the first few pages, when the script was several lines long. But perhaps there are people who haven't been following along who could indeed get the wrong impression that this is all SR. (that said, you were obviously just called out on a post you made months ago and simply defending that).

Maybe Dre can try a version at some point that keeps the rest of his script, but with the SR plugin code omitted. Even if the SR effect turns out to be minimal/nonexistent, he should be proud of the balance that he has struck. The thread has shifted away from interest in the SR technique itself, and most of the focus now is just on how good the results are. Same as any other project thread here. 

towne32 said:

Maybe Dre can try a version at some point that keeps the rest of his script, but with the SR plugin code omitted.

It's not possible as the SR plugin - while gaining (some) details from the adjacent frames - also cancels out noise and makes the image sharper too. With removing the plugin you'd remove the denoising and sharpening as well, leaving not much else but the anti-aliasing.