GOUT image stabilization - Released — Page 37

That was the thread of which I was thinking.  I searched for it for ages, but Frink is right: the search function is awful.  I can never make it return the desired results, even when I search for subjects that I know we have discussed at length. 

The last comparison definitely shows that the PAL is better in terms of detail.  I definitely think that the ROTJ script should be written for the PAL GOUT.

Chewtobacca said:

 

The last comparison definitely shows that the PAL is better in terms of detail.  I definitely think that the ROTJ script should be written for the PAL GOUT.

But you must have missed it, g-force only has the NTSC versions, so those will work, either way the finished product is going to look a TON better than the retail GOUT.

I guess we could convert our PAL version of Jedi to NTSC and then use it instead, if we wanted to do the extra work.

dark_jedi said:

But you must have missed it, g-force only has the NTSC versions, so those will work, either way the finished product is going to look a TON better than the retail GOUT.

No; I didn't miss it.  I am sure there is some way we could "enable" G-Force to work with the PAL ROTJ.

dark_jedi said:

I guess we could convert our PAL version of Jedi to NTSC and then use it instead, if we wanted to do the extra work.

That was my suggestion as well. :-)

vbangle said:

G-Force, is there anything you can share with us about your progress on the V3 ROTJ script? What are some of the time consuming tasks in tweaking your script for ROTJ that's different from the other two movies scripts?

You know, the most time consuming thing is really the convolution (deringing) filter. Painful really, getting rid of ringing without loosing too much sharpness. I suppose if I was working with that PAL version, I wouldn't even bother, as the ringing seems really low judging by the pics from the other thread. Then it's just a matter of getting the resizing and cropping right. Everything else in the script should just work, except the subtitles may need to be shifted by a few frames depending on if the PAL and NTSC are in sync or not.

-G

Doctor M said:

Probably a stupid question, but did you try dgdecod with the cpu > 4?

I've had pretty good success and it's fairly fast.

cpu: 0 to 6 (default: 0)
Post-Processing Quickset Options.
(Y=luma, C=chroma, H=horizontal, V=vertical)
- 0: DISABLE POST-PROCESSING
- 1: DEBLOCK_Y_H
- 2: DEBLOCK_Y_H, DEBLOCK_Y_V
- 3: DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H
- 4: DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H, DEBLOCK_C_V
- 5: DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H, DEBLOCK_C_V, DERING_Y
- 6: DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H, DEBLOCK_C_V, DERING_Y, DERING_C

You can use cpu2 if you want to disable/enable specific options (such as the strong deblocking).

Yes. the Deblocking in dgdecode in not necessary on these sources, and the de-ringing is not very effective and blurs quite a bit.

-G

g-force said:  You know, the most time consuming thing is really the convolution (deringing) filter. Painful really, getting rid of ringing without loosing too much sharpness. I suppose if I was working with that PAL version, I wouldn't even bother, as the ringing seems really low judging by the pics from the other thread.

That might mean that a ROTJ script based on the PAL GOUT would be less time consuming to write.  This would ease the burden on you somewhat.

Doctor M said:

One more suggestion: fft3dfilter includes a dering filter...

Too bad it doesn't work with our source either.

-G

Doctor M said:

Hey, just a quick OT thank you.

I was working on another project today and the source had some unpleasant aliasing.  I remembered a line in this script about it:

[code]########## anti-alias
NNEDI2(dh=true, field=0)
NNEDI2(dh=false,field=1)[/code]

I've never seen that done before, I'm not sure HOW it works, but it does a beautiful job.

Thanks.

 I'm glad it's useful even outside of the GOUT sources. You may not like the answer on how it works, but here goes anyway:

1. neural-network interpolate a line in between every line of resolution.

2. throw away all original lines and neural-network interpolate every line in between the first interpolated lines.

3. the next step that you didn't include is to resize back to original height.

That's right, there are absolutely no original lines of resolution remaining from the source. They are all interpolated. Kinda scarey I know, but it works amazingly well, doesn't it?

-G

That is scary.  So in effect we improve the GOUT by dispensing with the GOUT!

Chewtobacca said:

That is scary.  So in effect we improve the GOUT by dispensing with the GOUT!

 Good way to put it!

-G

g-force I know I asked you this a while back but I was wondering are you making any headway on the Jedi script or are you running into issues? I am curious to know if we might be able to get this bad boy set out by the Holidays is all, I am still currently working on the '97 SE v2 set now and probably have at least a couple weeks left before these are done, then I will have all my time on the v3 SW and ESB, then onto Jedi when ready.

Thanks

dark_jedi said:

g-force I know I asked you this a while back but I was wondering are you making any headway on the Jedi script or are you running into issues? I am curious to know if we might be able to get this bad boy set out by the Holidays is all, I am still currently working on the '97 SE v2 set now and probably have at least a couple weeks left before these are done, then I will have all my time on the v3 SW and ESB, then onto Jedi when ready.

Thanks

 With what I just learned about the PAL version, I am having difficulty finding the motivation to move forward with the NTSC version. This will have to wait until I get a PAL copy.

-G

Sent you a PM.

g-force said:

Doctor M said:

Hey, just a quick OT thank you.

I was working on another project today and the source had some unpleasant aliasing.  I remembered a line in this script about it:

[code]########## anti-alias
NNEDI2(dh=true, field=0)
NNEDI2(dh=false,field=1)[/code]

I've never seen that done before, I'm not sure HOW it works, but it does a beautiful job.

Thanks.

 I'm glad it's useful even outside of the GOUT sources. You may not like the answer on how it works, but here goes anyway:

1. neural-network interpolate a line in between every line of resolution.

2. throw away all original lines and neural-network interpolate every line in between the first interpolated lines.

3. the next step that you didn't include is to resize back to original height.

That's right, there are absolutely no original lines of resolution remaining from the source. They are all interpolated. Kinda scarey I know, but it works amazingly well, doesn't it?

-G

Haha, this is pretty awesome. I'm amazed at how advanced fan-based coding is becoming. I wonder how long until someone builds a freeware code to do the sort of dirt-replacement algoirthms that Lowry uses. In principle it doesn't seem like it is impossibly complicated...

g-force said:

 With what I just learned about the PAL version, I am having difficulty finding the motivation to move forward with the NTSC version. This will have to wait until I get a PAL copy.

-G

 Oh no!  I'm an NTSC guy, please don't quit!  :)

JediTray, G-Force is talking about using the PAL version because it is of higher quality.  A conversion to NTSC could be built into the script, so an NTSC release would still be made; it is just that the source would be PAL.

CompMovieGuy said:

I think I would rather IVTC in AVS then use DGIndex's Force Film since its very hit n miss. Although I have never tried it on the GOUT so it could work well in that instance.

You're completely right. It totally depends on source if DGIndex's Force Film works well or not. I've found it so unreliable that I gave up on it and since then I use manual IVTC (Decomb) which works damn fine.

I disagree with Doctor M, I don't know why you let yourself talk round so easily when you was so certain about it.