Interlace

[Kuronoe]

Hello again friends.

Can anyone help me with the interlace of this dvd?
I'm trying to restore the Mezzo SDA series and I can not get rid of the problems on those lines, especially in the mouth and around the characters.
I have tried all the combinations I know in the QTGMC, NNEDI3, IVTC and megui's suggestions (decomb and others), but none have worked.
I tried the rematch () which got slightly better, but also leaves some problems (artefacts and jagged lines).

PV: DROPBOX

IMGUR 01
IMGUR 02


Thank you in advance for your time.

[manono]

It's fieldblended. Try:

QTGMC()
Srestore(Frate=23.976)

It probably won't be perfect, but an improvement over an IVTC. You say you still saw interlacing after having used QTGMC. But that is almost, by definition, impossible. Perhaps you mean to say you saw aliasing.

[Kuronoe]

Thanks for the answer, Manono.

As I said, I tried many possibilities of the QTGMC, but did not solve until then. In the images I sent, I'm using QTGMC, you're right, it's working for the interlace and that problem is something else that I can not define.

I've tested some fieldblended and I'm testing XAA (), but the problem persists.

Thanks for the ideas.
And sorry for the inexperience: D

[jagabo]

Did you add SRestore as suggested?

[Kuronoe]

Did you add SRestore as suggested?

Yes, I do use these most of the time when I'm rip dvd.

In this print I'm using it. It is as if I did not interpolate the scene:

https://forum.videohelp.com/images/imgfiles/u4mJsMJ.png

[jagabo]

You're sample had almost no blending left after QTGMC().SRestore(). I don't think you're going to find anything better.

Mpeg2Source("Mezzo Forte (1)_track1_eng.d2v", CPU2="ooooxx", Info=3) 
QTGMC(preset="fast")
SRestore()

[Kuronoe]

Thank you, Jagabo.

I had come to this result, but I thought I would have a way of removing the other defects.
I'll try to look at a couple more sources, maybe it's the DVD that's the problem.
Apparently there is a blu-ray of that episode.
Thank you for your patience in helping me with these problems.

[jagabo]

maybe it's the DVD that's the problem.

Of course the DVD is the problem.

[Kuronoe]

Forgets.
The reviews on Amazon are not good. Apparently he has the same problems. I do not understand how they sell it.

Sorry for any silly comment.

[manono]

Thank you, Jagabo.

I had come to this result, but I thought I would have a way of removing the other defects.

Depending on how much work you want to put in, the remaining blends can be removed manually, either by using frame interpolation or frame duplication (if anime). Of course, this requires going over the whole thing manually to find the remaining blends.

It's also a good idea to check the field order before using SRestore. The sample unblended pretty well, but if the complete episode has a different field order it won't, and you might have to take that into account (AssumeTFF) before the bobbing/unblending.

[Kuronoe]

Depending on how much work you want to put in, the remaining blends can be removed manually, either by using frame interpolation or frame duplication (if anime). Of course, this requires going over the whole thing manually to find the remaining blends.

It's also a good idea to check the field order before using SRestore. The sample unblended pretty well, but if the complete episode has a different field order it won't, and you might have to take that into account (AssumeTFF) before the bobbing/unblending.

Hi, Manono.

I talked to a friend, he gave me the idea to encode normally and then remove/replace the roughest problematic frames with RemapSimples(). I'll think about it later. I deleted the remux I had made, I was discouraged after buying it.

I'll consider what you suggested when you go back to playing this dvd.

Thank you.

[jagabo]

Simple functions to replace a frame with a copy of the next or previous frame:

######################################################

function ReplaceFrameWithNext(clip Source, int N)
{
  # Replace frame at N with a copy of the frame after it

  loop(Source, 0, N, N)
  loop(last, 2, N, N)
}

######################################################

function ReplaceFrameWithPrev(clip Source, int N)
{
  # Replace frame at N with a copy of the frame before it

  loop(Source, 0, N, N)
  loop(last, 2, N-1, N-1)
}

######################################################

Sample call:

WhateverSource()
ReplaceFrameWithNext(1000) # replace frame 1000 with a copy of frame 1001
ReplaceFrameWithPrevious(2000) # replace frame 2000 with a copy of frame 1999

[jagabo]

And using RemapFrames to do something similar:

WhateverSource()
prevF = Loop(2,0,0).Subtitle("prev") # clip of previous frames
nextF = Trim(1,0).Subtitle("next") # clip of next frames

ReplaceFramesSimple(last, nextF, mappings="100 110 500") # replace frames 100, 110, 500 with frames 101, 111, 501
ReplaceFramesSimple(last, prevF, mappings="75 150") # replace frames 75, 150 with frames 75, 149

This is easier to use when you have lots frames to replace. You just add to the mappings strings.

[jagabo]

encode normally and then remove/replace the roughest problematic frames with RemapSimples()

That's not a good idea. Just replace frames in the same script:

WhateverSource()
QTGMC()
SRestore()

prevF = Loop(2,0,0).Subtitle("prev") # clip of previous frames
nextF = Trim(1,0).Subtitle("next") # clip of next frames

ReplaceFramesSimple(last, nextF, mappings="100 110 500") # replace frames 100, 110, 500 with frames 101, 111, 501
ReplaceFramesSimple(last, prevF, mappings="75 150") # replace frames 75, 150 with frames 75, 149

[Kuronoe]

Thanks for replying, Jagabo.

I was trying to use the "RemapFrames (mappings =" [1000 1001] 1003 "), but it does not seem to accept more frames, which is a problem.

The "ReplaceFrameWithNext()" is more efficient than the "RemapFrames()" above, but I also could not insert any more frames in it.

Your "ReplaceFramesSimple()" seems to work much better. I need to get time to test it. I only worry about the "trim" in the function, which qtgmc usually breaks with many cuts here.

Thanks for listening.
As soon as I test it, I'll tell you what it was like.

[jagabo]

I've used ReplaceFramesSimple with mappings of dozens of frames/ranges. ReplaceFrameWithNext/Prev can only replace a single frame. If you want to replace more you need to call ReplaceFrameWithNext for each frame you want to replace.

[Kuronoe]

I've used ReplaceFramesSimple with mappings of dozens of frames/ranges. ReplaceFrameWithNext/Prev can only replace a single frame. If you want to replace more you need to call ReplaceFrameWithNext for each frame you want to replace.

Hello, jagabo.

I had said that would give you a position on the outcome of the video.

So give up using QTGMC in this anime along with Srestore.
QTGMC works fine, but SRestore is problematic, as well as leaving some artifacts and some issues with the fieldblend.

Looking for other filters, I found the BruteIVTC, strangely it lined 80% of the frames that were problematic and adding the Tdecimate improved by 85% overall.

BruteIVTC ()
Tdecimate (rate = 29.970, mode = 7)

I did not understand how this was possible, in fact it was luckier than technique.
After that I just had to sort out random frames using the script you provided above.

I'm still testing, but I wanted to pass this information on. Maybe you have an idea of how this worked.


Thanks in advance.

[jagabo]

I tried BruteIVTC and it left lots of comb artifacts. It also screws up the chroma channels on interlaced YV12 sources. Use ConvertToYUY2(interlaced=true) before calling it.

You need to TDecimate() to 23.976 fps (or maybe 25). Otherwise you will get jerky pans.

So if you want to use BruteIVTC() and TDecimate() try something like:

Mpeg2Source("Mezzo Forte (1)_track1_eng.d2v", CPU2="ooooxx", Info=3) 
AssumeTFF()
ConvertToYUY2(interlaced=true)
BruteIVTC()
Tdecimate(rate = 23.976, mode = 7)
ConvertToYV12()
vinverse() # blend away residual combing

[Kuronoe]

Hello, Jagabo.

Yes, the problem with the chroma is complicated to solve, I did not find a better way to solve beyond those suggested.

I was mapping the most problematic scenes by replacing the "next" and "previous" using the script that you provided, but some have no way to combine, they are different.

I used the Tdecimate in doubt, not the fps change, but somehow mode = 7 seems to have removed some problems here. I left the 30fps to avoid the "jumps" in panoramic scenes, the ones that move to show the scenery. However, in these scenes I can not map because they "jump" as much with 24fps as 30fps.

The above "pv scene" I had to sort out some parts manually, like the one around the head of the young woman, but I have doubts whether it is feasible so much service by a eccentricity.

I still want to test other things, with qtgmc some scenes look better than with BruteIVTC, I was thinking of replacing each other instead of mapping many frames, but I do not know how to do it. The idea was:

#video 01
FilterMap1 = "[ranger ranger] [ranger ranger]"
FilterLine1 = QTGMC (preset = "medium", fpsdivisor = 2, others)
ReplaceFramesSimple (FilterLine1, mappings = FilterMap1)

#video 02
FilterMap2 = "[ranger ranger] [ranger ranger]"
FilterLine2 = BruteIVTC (). Tdecimate (29.970, mode = 7)
ReplaceFramesSimple (FilterLine2, mappings = FilterMap2)

But I do not know if it will work.
I'll try it with more time on the weekend.


Anyway, thanks for the help.

[danfgtn]

I tried BruteIVTC and it left lots of comb artifacts. It also screws up the chroma channels on interlaced YV12 sources. Use ConvertToYUY2(interlaced=true) before calling it.[/CODE]

When following the recommendation to use ConvertToYUY2(interlaced=true) before BruteIVTC, I have a problem with the image, it seems to generate a "chromableeding" (i think) in red color, which does not happen if I remove ConvertToYUY2.

Look at this image, the red seems to "bleed"(?) on the sleeve of the shirt:
https://diff.pics/13KAUuMKSYLT/1

[jagabo]

I believe the rougher color edge in the YUY2 image is actually the correct interpretation. And the SRC image looks smoother because the chroma from the two fields is blurred together.

[jagabo]

By the way, I tried the BruteIVTC filter with some cleaner telecined film and it didn't suffer from the problems I was seeing with your sample clip. So it appears to work properly with interlaced YV12 -- and the problems I was seeing had to do with the residual combing from the blended fields in your source.