Can't get Interframe to work to convert PAL to NTSC

[jagabo]

The biggest problems I saw were at scene changes. You need to change the scene change thresholds for those. Other problems were typical of what happens when the motion interpolation doesn't work well because the shot is too blurry and/or the motions too large.

[VideoFanatic]

Do you have a script with the correct scene change settings please? Also at the end of the video you can see the advertisement is all garbled as well.

[jagabo]

Do you have a script with the correct scene change settings please?

I'm sure you can read the instructions as well as I can.

Also at the end of the video you can see the advertisement is all garbled as well.

Typical motion interpolation artifacts. As johnmeyer mentioned, you can try playing with block size and overlap values. But I don't think you'll get much better results with the ad.

[poisondeathray]

You're talking about the occlusion artifacts typical of motion interpolation. Interframe has them too, but using "smooth" hides many of them. The "cost" is the image is more blurry , and you get "strobing" in some scenes

To reduce the artifacts in mflowfps, use a larger blocksize (e.g. blksize=32) , you'll get about the same artifacts as interframe using "smooth", but the results will be sharper. But one thing interframe using smooth preset handles well are the static overlays like the WWE Live logo. The motion underneath can confuse interpolation algorithms. There might be some settings you can tweak in mflowfps for that

EDIT - that's one thing the interframe smooth preset does really well - it handles the overlays nicely



If you want absolutely zero artifacts, don't use motion interpolation. Use ChangeFPS(60000,1001) to duplicate frames. The motion won't be as smooth, but you'll have zero of the occlusion artifacts such has interframe or mflowfps have

[jagabo]

When I originally downloaded your sample video the sources wasn't there. After noticing you had added the source I downloaded it and ran your script (aside from some minor cropping/addborder changes at the end). I'm getting much cleaner results than you.

ffVideoSource("2 minute Source.mkv") 
AssumeTFF()
Spline36Resize(1280,height).AddBorders(0,0,16,0)
QTGMC("Super Fast", SubPel=2) #SubPel must match pel in MSuper below
super = MSuper(levels=1, pel=2, hpad=16, vpad=16) #For very fast and quicker QTGMC presets
MFlowFps(super, QTGMC_bVec1, QTGMC_fVec1, num=60000, den=1001)
Crop(0,0,-16,-0)
Spline36Resize(1280,720)

Overall, I'm getting more blending (which is much less visible at normal playback speed) instead of distortion.

[VideoFanatic]

I don't know if this made a difference but I cut the artifacts sample out of a 3 hour video. Don't know if running my original script would then create the same artifacts in the same places in the 2 minute sample I took from that 3 hour video? I'll try later on today.

Is that correct that you've got 2 QTGMC lines?

I don't suppose you know how to get Interframe working do you? It never encodes a full video when I encode HD material. I've got the latest interframe extension files.

[jagabo]

Is that correct that you've got 2 QTGMC lines?

Sorry. No, I didn't have two calls to QTGMC(). I made a mistake when I cut/pasted from my script. I'll fix the post.

I don't suppose you know how to get Inframe working do you? It never encodes a full video when I encode HD material. I've got the latest interframe extension files.

Interframe? It works fine here (though I have an old version of SVPflow.dll). But I don't have a Nvidia GPU so I have to use GPU=false.

[VideoFanatic]

Does using a slower QTGMC setting in the MFlowFPS script give less artifacts or does the speed not make a difference?

[jagabo]

I think your artifacts are from MFlowFPS, not QTGMC. But why don't you try it.

[johnmeyer]

I played the clip using VLC at 0.2x speed. I see the usual artifacts in exactly the places I would expect them. Motion estimation does not like strong vertical lines, so any time something crosses in front of the referee's vertically striped shirt, you see the bars on his shirt break apart. Even worse are the vertical shades behind the crowd (they look a little like the pipes in a pipe organ). The artifacting on those is really quite strong.

The artifacts on scene changes can be easily fixed by following jagabo's advice, but these others can merely be tuned a little.

The "ultimate" way to do these conversions is to also do a conversion simply using a blend (see advice given earlier). Then, in your NLE, switch to the blended version when the ME breaks down. If someone were paying me for this, I'd go one step further and use a mask for the frames where the ME breaks down, and only let through enough of the blended frame to cover the place where the motion breaks down.

[johnmeyer]

I re-read both my own posts and those of poison and jagabo, and followed our collective advice. Using the script I linked to, modified as I already recommended, I got rid of all the scene change artifacts, and reduced the others to the point where they aren't going to be noticed. The result looks much, much better than what you posted.

I'll upload the result when I get a chance. You will still need to adjust the script so that the levels in the final match your original. As part of this, you may not need the ConvertToYUY2 line. Also, 720x480 square pixel may not give you the correct aspect ratio. You may have to work on that.

Here's the script I used:

#Script to convert PAL to NTSC

loadplugin("C:\Program Files\AviSynth 2.5\plugins\MVTools\mvtools2.dll")
loadplugin("C:\Program Files\AviSynth 2.5\plugins\RemoveGrainSSE2.dll")
loadplugin("C:\Program Files\AviSynth 2.5\plugins\SSE3Tools.dll")
loadplugin("C:\Program Files\AviSynth 2.5\plugins\mt_masktools-25.dll")

Import("E:\Documents\My Videos\AVISynth Scripts\QTGMC-3.31a.avsi") #Use version with hpad,vpad hardwired to 8 (avoids crash)

SetMemoryMax(1000)
SetMTMode(5,3)
AviSource("e:\fs.avi").AssumeTFF()
SetMTMode(2)

ConvertToYUY2(interlaced=true, matrix="PC.601")

QTGMC( Preset="Slower", SubPel=2, EdiThreads=4,ThSCD1=180, ThSCD2=98, Blocksize=32,Overlap=4)  #Nick's latest settings default scene detection is thSCD1=304, thSCD2=100
super = MSuper(levels=1, pel=2)
MFlowFps(super, QTGMC_bVec1, QTGMC_fVec1, num=60000, den=1001,ThSCD1=180, ThSCD2=98)
Spline36Resize(720,480)

blur(0.0,1.0)
sharpen(0.0,0.75) #Adjust 2nd value between 0.5 and 1.0 to taste

P.S. Someone who knows more about the scene change logic may want to suggest better ThSCD1 and ThSCD2 parameters.

[johnmeyer]

Here's a link to the file that results from the script in my previous post. Everyone has their own standards, but to me it looks pretty darned good. Your result had artifacts at most of the scene changes, and also had breakdowns every time an object crossed in front of anything with a strong vertical line. By contrast, the scene change logic worked as it should, once the settings were changed the way I recommended, and you now get a single blended frame at the scene transition, which is what has to be done in order to maintain the 59.94 fps cadence, but at the same time avoid trying to create a motion-estimated frame which, as you saw in your results, becomes a bizarre mess.

What surprised me was how well it worked once I increased the block size. Since the same vectors are used for both QTGMC and also for the subsequent MFlowFPS call, this increased block size affects both operations.

PAL 50i to NTSC 60p

This file is 720x480, suitable for burning directly to DVD. It is designed to be used by a player that understands how to stretch square pixels, and is designed to be played at 1.2121 (NTSC DV Widescreen) PAR.

[johnmeyer]

I guess there is no guarantee that anyone who starts a thread will stick around and acknowledge the help they got. However, the lack of response from the OP is puzzling to me in this instance because he sent me a PM specifically asking me to help him. I then provided both a script and a final result that seems to have solved all the problems he reported (my result is infinitely better than the result he posted). Perhaps he got called away to other things ...

[VideoFanatic]

It'll take me 2 days to encode my full length video. I can't report back the results until that's done.

[johnmeyer]

You should never encode a full video until you have first tested any restoration technique on a few short (5-15 second) clips. If you don't know the results prior to doing a "2-day encode," you are, very literally, wasting your time.

[VideoFanatic]

I've already done that. But what I see on a 2 minute clip and a 3 hour video are 2 different things.

[johnmeyer]

I've already done that. But what I see on a 2 minute clip and a 3 hour video are 2 different things.

???

In what way are they different? Are you saying that the quality changes somehow when you encode a longer clip? I've sure never seen that before.

[VideoFanatic]

Just because there are no issues on a 2 minute clip doesn't mean there won't be any issues on a 3 hour video (excluding the 2 minute portion).

You're a strange guy John. In the past you might make 1 post in some threads of mine then I never hear from you again when I reply. That was why I emailed you as perhaps you didn't subscribe to the thread. Now I can't stop you posting! I would rather reply to the thread when I have something of value to report, I don't think anyone wants to read our posts discussing insignificant things like why I'm not posting, etc.

For the record, you CAN get different results when encoding a short clip versus the whole video. I tried that once with 720p 50fps QTGMC Super Fast at around CRF 20 on a short clip of a man walking to an arena. CRF 20 produces pixellation which can be smoothed out with QTGMC. The QTGMC Super Fast clip produced a less pixellated picture on the clip. But the full length video at 2 pass was even less pixellated.

[johnmeyer]

You're a strange guy John. In the past you might make 1 post in some threads of mine then I never hear from you again when I reply. That was why I emailed you as perhaps you didn't subscribe to the thread. Now I can't stop you posting! I would rather reply to the thread when I have something of value to report, I don't think anyone wants to read our posts discussing insignificant things like why I'm not posting, etc.

I just looked at every thread in this forum in which we both posted during the past two years. Not once did I ever fail to respond to any post you made, and I never once left you hanging. Here is the list of those posts:

https://forum.videohelp.com/threads/379618-Can-t-get-Interframe-to-work-to-convert-PAL-to-NTSC
https://forum.videohelp.com/threads/375556-QTGMC-EZDenoise-Help-Needed
https://forum.videohelp.com/threads/355012-Is-vReveal-any-good-for-video-restoration
https://forum.videohelp.com/threads/375183-How-to-Fix-Crushed-Blacks
https://forum.videohelp.com/threads/375091-4-3-video-in-16-9-frame
https://forum.videohelp.com/threads/375002-Cropping-2-35-Movie-to-Enlarge-the-Height/page2
https://forum.videohelp.com/threads/374435-How-to-Convert-AVIs-to-DVD-And-Avoiding-buying-AVI-DVDs

If anyone cares to read those threads, just use your browser search to find my user name and see what I wrote. I think my answers were timely, thoughtful, helpful, and fairly detailed and complete.

So, your statement is flat wrong.

However, I have a simple solution that will ensure that what never happened will never happen again ...

[VideoFanatic]

I wasn't accusing you of anything bad. People are free to help or not help as they wish. I'll take your word for it John, I'm not going to waste time reading posts about an insignificant thing like I mentioned before. I apologize for confusing you with someone else.

[manono]

This is slightly modified from something jagabo offered once:

super=SVSuper("{gpu:0}")###unless you have a compatible GPU
vectors=SVAnalyse(super, "{}")
SVSmoothFps(super, vectors, "{rate:{num:20, den:18}}", url="www.svp-team.com", mt=1)

Fill in the ratio of what you want in num and den (like 60000 and 1001?) and check if you've gotten the correct framerate before encoding. It needs both SVPFlow1.dll and SVPFlow2.dll. I haven't used it in a while, but as I remember it's quite fast.

Me, I don't use any of that interpolation stuff when converting interlaced material, but do it the old fashioned ChangeFPS way.

[VideoFanatic]

Sorry for the delay. Been having issues with my PC.

JohnMeyer - I tried your script and it causes the picture to jump up every several seconds or so. Also tried the alternative settings that were commented out.

Manono - I tried your script but don't know how to get it to output 60fps. The documentation says this:

num: 2 = Numerator of multiplier for frame rate change.

den: 1 = Denominator of multiplier for frame rate change.

But it doesn't explain how to work out the multipliers. What settings should I use for 60fps when QTGMC is running before your script?

Regarding the Interframe script I posted in my 1st post. Open further testing on my 2 hour video, the script works when encoding 1 video but when you try to encode 4 videos at once, all of them will fail or all but 1 will fail to encode. Encoding completes for all 4 but before it completed I noticed 5 minutes before the encoding completes that the projected file size is nowhere near what it should be. When I play the encodings back, the video just pauses after 10 minutes but you can still hear the audio. So it seems that Interframe doesn't work when used with QTGMC when encoding multiple videos at once. However using Yadif instead of QTGMC works fine for encoding multiple videos at once. I'm NOT using MT.

Also I can't see any artifacts on the 2 minute source sample when using the Interframe script and it doesn't look blurry to me. So interframe seems to be the best script at the moment unless anyone has anything better?

[manono]

Manono - I tried your script but don't know how to get it to output 60fps.

After you have 50fps and you want to interpolate to 59.94fps, this seems to work:

super=SVSuper("{gpu:0}")###unless you have a compatible GPU
vectors=SVAnalyse(super, "{}")
SVSmoothFps(super, vectors, "{rate:{num:1200, den:1001}}", url="www.svp-team.com", mt=1)

[johnmeyer]

JohnMeyer - I tried your script and it causes the picture to jump up every several seconds or so. Also tried the alternative settings that were commented out.

It does?? I don't even know how it could do that. If you download the version that I uploaded for you, which was created by this script, can you give me a few frame numbers where you see some sort of jump?

[VideoFanatic]

31 secs and every several seconds or so after that. This is my script:

Spline36Resize(1280,height).AddBorders(0,0,16,0)
QTGMC( Preset="Super Fast", SubPel=2, ThSCD1=304, ThSCD2=100, Blocksize=32,Overlap=4)  
super = MSuper(levels=1, pel=2)
MFlowFps(super, QTGMC_bVec1, QTGMC_fVec1, num=60000, den=1001,ThSCD1=180, ThSCD2=98)
Crop(0,0,-16,-0)
Spline36Resize(1280,720)
blur(0.0,1.0)
sharpen(0.0,0.75)

Strangely your encoded version doesn't have the screen jumping up problem. However I can see glitches in the channel logos. So even if I get your script working properly, my interframe script still gives better results at the moment as there's no glitches at all.

Manono - the same glitches are in your script as well.

[manono]

Manono - the same glitches are in your script as well.

Okay then. Sorry for having wasted your time.

[poisondeathray]

my interframe script still gives better results at the moment as there's no glitches at all.

No glitches ? That's a huge overstatement . There are artifacts using interframe as well. However, the overlays are definitely cleaner with interframe (but still not perfect), because you're using the smooth preset - at the expense of blurring and more loss of details. With GPU on there's glitches on the frame borders, CPU mode is better , cleaner and more stable, albeit slower

The only glitch free way is without optical flow methods (changefps or convertfps), or if you manually make adjustments / masking in other programs

People might get slightly different results depending on the specific dll's being used, but interframe definitely has glitches and artifacts - you're deluding yourself if you think there are none

[VideoFanatic]

Can you give me some frame numbers where you see glitches in the interframe video?

[poisondeathray]

Can you give me some frame numbers where you see glitches in the interframe video?

Is that a serious question ? Motion interpolation is never "perfect". (There is no "perfect" way to do this conversion)

What type of glitches did you want frame numbers on ? Just look at the video. Look at the overlay glitches, or border glitches with GPU=true, look in parts whenever there is big motion - there are always some artifacts and glitches. It's just a matter if the problems are "acceptable enough", or perhaps the problems might not be clearly visible at 60FPS speed

An additional problem with interframe using smooth is the strobing motion , blurriness. ConvertFPS would be cleaner but still have that strobey blended frame look, but without the glitches. ChangeFPS would be the cleanest, but motion not as smooth

[jagabo]

If you're talking about the video in post #30?

1-14, the WLive logo and wrestler's boot in the lower right corner.
12-15, the referee's shirt.
43-49, Sky Sports logo in the top right corner
100-101, everywhere
478-905, the vertical blue/black stripes in the background

The list goes on and on. Virtually every frame has some glitches and blending if you look closely, though many aren't really visible at normal playback speed.