[Help] Terminator 2 Judgment Day UAR

[CourtlyHades296]

I'd like to see the screenshots uploaded to ultimateaspectratio website.

As for the D-Theater, buy a copy and scan it.

[spoRv]

Further tests are needed; I'll eventually post them there when it will be the time.

D-Theater: I have a lot of hardware, but sadly not a D-VHS deck...

About WMV: found a comparison here http://www.dvdbeaver.com/film/dvdcompare/term2.htm
It has the same color grading as blu-ray (disreguard the tiny color difference, given by the fact the WMV is rec601 and BD is rec709, and the original poster didn't the conversion - because he thought it wasn't the case, as it should have been, as WMV is HD, albeit "only" 1440x816 instead 1920x816); do note that, strangely, the resolution is higher than BD...

[spoRv]

Screenshot comparisons - UK PAL laserdisc widescreen Vs UAR (using the former and Russian DVD fullscreen):

http://ultimateaspectratio.ga/Terminator...79804.html
http://ultimateaspectratio.ga/Terminator...82422.html
http://ultimateaspectratio.ga/Terminator...83013.html

[CourtlyHades296]

Is there any footage of this in motion?

[spoRv]

Not yet; still encoding Robocop UAR since friday, and still another one day or two left!

When finished, I'll try to make a test clip.

[random.next]

@spoRv
If you have that LD captured please share to pm. I want to make a custom open matte (16:9) version with good alignment. With BD it was really difficult.

[spoRv]

I've just captured few minutes, to make rough tests; I'll capture the whole film in the next few days. But lossless capture will take around 40/50GB, so I'd prefer to make a compressed version - tell me, which is a good compromise for you, considering it is SD, with active area around 700x300 pixels?

About alignment: I made only a brief clip, with Sarah and the doctor as hostage; it works usually well, but sometimes it goes out of control during the fastest horizontal pans.

As a shot usually is longer than few frames, I thought about an option for an eventual script revision, that tells it something like:
"if the shot aligns N frames with same (or similar) settings, then alignment settings suddenly change for C frames (user given variable, let's say 3 or 4 as default) more than a given percentage, then return to the same alignment of N... consider them out of alignment, and align those C frames using N settings"

Got the point? This usually happened before with explosions, or other nearly one color frames, which the script, of course, could not know how to align. And, in this case, during the fastest pans, where again the frames have almost one color, and script get confused.

[random.next]

Lossless is good for me but if it is too large for sharing I'd prefer minimum comperssion as possible especially for SD resoultion (output QP not higher than 10).

>As a shot usually is longer than few frames, I thought about an option for an eventual script revision, that tells it something like:
"if the shot aligns N frames with same (or similar) settings, then alignment settings suddenly change for C frames (user given variable, let's say 3 or 4 as default) more than a given percentage, then return to the same alignment of N... consider them out of alignment, and align those C frames using N settings"

You are able to control that through maxDiff and maxDiffIncrease args. Explosions and other bright shots will increase DIFF value because of higher dynamic range than dark scenes. Try increase maxDiffIncrease from default 1 to 2-5 e.g. How these parameters work briefly explained in the description. If you have questions ask them. High values of maxDiff and maxDiffIncrease (max is 255) will cause that all frames will be included to one scene.

But the best way is to use statFile and editor as I already said many times before. It's really save your time

Anyway if this LD is really best for alignment i.e. have static overlay settings during the scenes then I can just give you my statFile and you can use it with different render.

[spoRv]

Lossless is good for me but if it is too large for sharing I'd prefer minimum comperssion as possible especially for SD resoultion (output QP not higher than 10).

Let me first capture it, then we'll see how big the lossless file will be.

You are able to control that through maxDiff and maxDiffIncrease args. Explosions and other bright shots will increase DIFF value because of higher dynamic range than dark scenes. Try increase maxDiffIncrease from default 1 to 2-5 e.g. How these parameters work briefly explained in the description. If you have questions ask them.

Thanks, I'll do some tests ASAP.

High values of maxDiff and maxDiffIncrease (max is 255) will cause that all frames will be included to one scene.

Please explain.

But the best way is to use statFile and editor as I already said many times before. It's really save your time

You are right, you told me a lot of times to use the stat file... still, I don't get how to use it... I mean, should I make one pass to let the stat file to be created? If so, how could I read it, as the instructions tells only to save that file, and not how to read it?

Anyway if this LD is really best for alignment i.e. have static overlay settings during the scenes then I can just give you my statFile and you can use it with different render.

What if we join our forces, and collaborate to make two editions? One will be UAR (mine), the others open matte a.k.a. "ambilight UAR" (yours) .

[random.next]

>Please explain.
With maxDiff=255 and maxDiffIncrease=255 all frames will have the same overlay params detected at first frames (backwardFrames + forwardFrames).
So if you know that some scene have fixed framing use special engine for it.

Code:

default=OverlayEngine(...)
fixed=OverlayEngine(maxDiff=255, maxDiffIncrease=255...)
default.Trim(0,100) + fixed.Trim(101,200) + default(201,10000) + fixed.Trim(10001,20000)...

This is not good usecase. Better is to use editor especially if most scenes have fixed framing (sources are stablized to each other). But combining engines with different settings is sometime useful.

>You are right, you told me a lot of times to use the stat file... still, I don't get how to use it... I mean, should I make one pass to let the stat file to be created? If so, how could I read it, as the instructions tells only to save that file, and not how to read it?

Steps:
1. Choose best OverlayConfig for given sources. No any stat at this step, just test different params at many scenes. To analyse the align quality set OverlayRender.mode=7. This is difference mode. If clips aligned perfectly then intersection area will be pure grey (or black in old avisynth). OverlayRender.debug=true is also useful. Your target is to minimize diff value. First of all you need to check:

• aspect ratio and angles params 
  
• sourceArea and overlayArea params
  
• static color correction if needed
  
• logo masks if needed
  
• are sources stablized to each other or not. If not then set backwardFrames=0.
  

I have some ideas how to support unstablized clips better and much faster but had no time to implement. So maybe this T2 LD will be good testcase.
Recommendation: use multithreaded resizing filters such as resampleMT: https://github.com/jpsdr/ResampleMT/releases
You need to specify all custom resize filters explicitly by downsize and upsize params. It increases alignment prefomance greatly.

2. Run analysis pass. Render is not needed at all. Disable all heavy filters except that needed for engine. Example:

Code:

OM=AviSource("c:\test\OpenMatte.avs") # YV12 clip
WS=AviSource("c:\test\Widescreen.avs") # YV12 clip
config=OverlayConfig(aspectRatio1=2.3, aspectRatio2=2.5)
OverlayEngine(OM.ConvertToY8(), WS.ConvertToY8(), configs=config, statFile="c:\test\Overlay.stat", editor=false)

For stablized sources it takes about few hours. 
Analysis pass in terms of avisynth is the playing full video without recording output for writing some technical info by some filters. This info is further repeatly used without recalculation. Once you have specified statFile align settings will be written to it and reused by this engine in the future. 
>If so, how could I read it, as the instructions tells only to save that file, and not how to read it?
It will be read by the same engine after analysis pass. No additional actions are necessary. Advice: backup statFile after analysis.

3. Run editor by OverlayEngine.editor=true. Check alignment in the every scene (scene in this case is the frame sequence with equal align params). I usually check the first and last frames in every scene. It is easy to correct bad frames by including frame to the other scene. Or using autooverlay and pan&scan buttons. All changes will be save in statFile and used them by engine.

4. Finally add OverlayRender to the script and encode it. 

> What if we join our forces, and collaborate to make two editions? One will be UAR (mine), the others open matte a.k.a. "ambilight UAR" (yours) .
The main task is to make good statFile. I will try if sources are compatable to each other. All other things are project specific.