logo Sign In

Star Wars GOUT in HD using super resolution algorithm (* unfinished project *) — Page 2

Author
Time

You appear to have pasted the same address twice.

Author
Time

Sorry, the link has been replaced.

Author
Time

Although I understand most of your points, I have to disagree with the idea that smart upscaling super resolution does not work. As a scientist I like to stick to the scientific evidence, and the developers of this super resolution algorithm have pretty convincing numbers on their site that prove that their algorithm is superior to many other methods described in scientific literature:

http://www.infognition.com/articles/video_resize_shootout.html

While I agree that the Despecialized Edition of Star Wars is amazing, especially v2.5, the quality difference in some of the elements in TESB and ROTJ is quite noticable. I'm sure Harmy is working on bringing those up to the quality of SW, but these despecialized films are many years in the making. By the time he is finished we may actually have seen the official bluray release of the originals. In the mean time I'm simply interested in what these super resolution algorithms can do. They keep getting better, and better so who knows we may be able to squeeze some more details from the GOUT.

I'm aware of the great work done by others to improve the GOUT in many ways, but personally, although I'm no fan of film grain, I prefer the grain over less detail. So out of personal interest I will continue to check out these algorithms, and anyone who's interested can follow my progress...  

Author
Time

For the full quality I suggest downloading the videos.

Author
Time

My two cents...

After comparing to the despecialized edition, I can see Harmy's version has much more clarity and less grain. I did like the HD GOUT's color though. The halls of the Tantive were much brighter and cleaner.

I look forward to seeing how this progresses.

-Ron

Author
Time

I will try to put it up asap. But what's wrong with frame 100? I can't see it.

Author
Time

It does seem like you've gotten a bit more real detail out of the GOUT, and when I color corrected it to match Team Blu's latest version, it seems that in terms of real detail, yours has the edge in several respects. The Team Blu version tends to smooth out some detail while exaggerating it in other places, whereas your upscale reveals detail without the distortion common to sharpening. However I did notice some aliasing issues in the video, where a smooth gradient in the GOUT may be noticeably more jagged and pixelated in your upscale. I don't know if it's an encoding issue or if it's in the raw file.

You probably don’t recognize me because of the red arm.
Episode 9 Rewrite, The Starlight Project (Released!) and ANH Technicolor Project (Released!)

Author
Time

What NeverarGreat told is true; take a look for example at the two pinkish tubes on the left over the stormtrooper's shoulder, at Leia's hair, both troopers "eyes" and "noses"...

Sadly my projects are lost due to an HDD crash… 😦 | [Fundamental Collection] thread | blog.spoRv.com | fan preservation forum: fanres.com

Author
Time
 (Edited)

Here's another sample:

Update: Clips removed to make space for new clips.

Author
Time

...and a first sample for The Empire Strikes Back:

Standard definition:

https://drive.google.com/file/d/0B8_LYKyZDiajY194Wm9wblFPSFE/view?usp=sharing

Upscaled:

https://drive.google.com/file/d/0B8_LYKyZDiajeFA0VkZuaENiY1U/view?usp=sharing

Author
Time

Don't know if this is due to compression artifacts (even if I don't think so, because 20Mbps should be enough to avoid them), but it seems that the superresolution algorhythm you are using is not that effective... a simple avisynth bicubic upscale, using the compressed MKV SD video you provided, sharpened a bit, seems very similar...

http://screenshotcomparison.com/comparison/117604

what do you think?

Sadly my projects are lost due to an HDD crash… 😦 | [Fundamental Collection] thread | blog.spoRv.com | fan preservation forum: fanres.com

Author
Time

_,,,^..^,,,_ said:

Don't know if this is due to compression artifacts (even if I don't think so, because 20Mbps should be enough to avoid them), but it seems that the superresolution algorhythm you are using is not that effective... a simple avisynth bicubic upscale, using the compressed MKV SD video you provided, sharpened a bit, seems very similar...

http://screenshotcomparison.com/comparison/117604

what do you think?

 To me it looks like it has some standard up-scaling artifacts the way you do it. The superres one looks softer but cleaner, and generally more appealing.

Author
Time

Wazzles said:

To me it looks like it has some standard up-scaling artifacts the way you do it. The superres one looks softer but cleaner, and generally more appealing.

Indeed, I used one of the most simple upscaling filter, but have you noted the problems on the superresolution version? OK, I must admit they are very tiny - infact you have to zoom in a lot to see them - but they are there... look in particular the eyes...

I expected much more quality, according to comparison examples...

Sadly my projects are lost due to an HDD crash… 😦 | [Fundamental Collection] thread | blog.spoRv.com | fan preservation forum: fanres.com

Author
Time

_,,,^..^,,,_ said:

Wazzles said:

To me it looks like it has some standard up-scaling artifacts the way you do it. The superres one looks softer but cleaner, and generally more appealing.

Indeed, I used one of the most simple upscaling filter, but have you noted the problems on the superresolution version? OK, I must admit they are very tiny - infact you have to zoom in a lot to see them - but they are there... look in particular the eyes...

I expected much more quality, according to comparison examples...

 What I saw look like interlacing and aliasing. The aliasing is part of the source, but the interlacing-esque artifacts were strange.

Author
Time

Which software do you use to get the individual frames? The raw video is a bit too large to put on google drive.

Author
Time
 (Edited)

_,,,^..^,,,_ said:

I expected much more quality, according to comparison examples...

I can imagine the results would work much better on something that isn't the GOUT, since this algorithm relies on blending multiple frames together to get added detail, whereas the GOUT has had multiple frames blended together to remove detail...

Perhaps we could test it out on the Technidisc? I notice in the example shot on Wikipedia, the processed image has a lot less rainbowing.

This signature uses Markdown syntax, which makes it easy to add formatting like italics, bold, and lists:

Author
Time

It will be great to see some examples using very good DVD, and compare the result of superresolution with the same title in BD (that, of course, should be of high quality, too).

Sadly my projects are lost due to an HDD crash… 😦 | [Fundamental Collection] thread | blog.spoRv.com | fan preservation forum: fanres.com

Author
Time

I will convert a scene from the 2004 dvds to compare with the 2011 blurays.

Author
Time
 (Edited)

_,,,^..^,,,_ said:

Don't know if this is due to compression artifacts (even if I don't think so, because 20Mbps should be enough to avoid them), but it seems that the superresolution algorhythm you are using is not that effective... a simple avisynth bicubic upscale, using the compressed MKV SD video you provided, sharpened a bit, seems very similar...

http://screenshotcomparison.com/comparison/117604

what do you think?

 

here's a better comparison.

http://screenshotcomparison.com/comparison/117784

Since the artifacts look more like interlacing problems then blown up interlacing problems, eedi takes care of it easy enough. I also added a sharpening filter to try and match yours. Now the difference is incredibly subtle, I'd guess it's temporally finding noise and dithering it into smaller noise?.. though without a raw frame its hard to say, fake fine grain + recompression might look the same

edit: ignore any differences in sharpening

Author
Time

[standard comment about imgur images being automatically compressed]

Author
Time
 (Edited)

I added a fair comparison: precise bicubic (A = -0.75) vs super resolution

The differences are subtle, but you can see more detail in the eyes, eye lashes, hair in the super resolution one. You could try sharpening the bicubic, but this would also increase the grain/noise.

http://screenshotcomparison.com/comparison/118152

Author
Time

DrDre said:

I added a fair comparison: precise bicubic (A = -0.75) vs super resolution

The differences are subtle, but you can see more detail in the eyes, eye lashes, hair in the super resolution one. You could try sharpening the bicubic, but this would also increase the grain/noise.

http://screenshotcomparison.com/comparison/118152

 Can you add the untouched GOUT frame for reference as well?

Author
Time

DrDre said:

I added a fair comparison: precise bicubic (A = -0.75) vs super resolution

The differences are subtle, but you can see more detail in the eyes, eye lashes, hair in the super resolution one. You could try sharpening the bicubic, but this would also increase the grain/noise.

http://screenshotcomparison.com/comparison/118152

 I still notice pixelation around Leia's mouth.

You probably don’t recognize me because of the red arm.
Episode 9 Rewrite, The Starlight Project (Released!) and ANH Technicolor Project (Released!)