De-interlacing - Is My Workflow Right?

[hizzy7]

I am getting closer with Removespots(), but it keeps saying "unable to load RemovegrainTSSE3, even though it is in my plugins folder. lol!!

[jagabo]

Does your CPU have SSE3 extensions? If not, try RemoveGrainSSE2.dll. Actually, just forget SSE3 and use SSE2.

[hizzy7]

Hi Friends!

I would like to know if I have correctly diagnosed this crappy video:

http://files.videohelp.com/u/183506/JU%20JITSU.demuxed.m2v

I think that is a poor PAL to NTSC transfer and that I should use this script:

Mpeg2Source("VTS_01_1.d2v")
AssumeTFF().QTGMC(Preset="placebo")
SRestore()
AssumeFPS(23.976)

Am I right?

Thanks again for your help!

[jagabo]

No, I think it was 29.97p.

[hizzy7]

Hi!

I have been playing with it and find the video confusing. If I may ask, what is the cause of all of the combing?

Thanks for your help!!

h

[jagabo]

There are a few shots where the video was sped up or slowed down for effect. But the base video was shot at 29.97p. Add frate=29.97 to SRestore() and remove the AssumeFPS().

[manono]

No, I think the source was 29.97p.

It depends on the scene. In some the framerate is greater even than 29.97fps such as the action scenes in the first part, perhaps to make the action seem faster than it really is. Much of it is 29.97fps. Not sure how to handle it overall, though. Is this for DVD?

Edit: I didn't see your additions before posting. If you make the whole thing progressive at 29.97fps, you'll get a lot of missing frames. He could unblend it to 59.94fps followed by reinterlacing it.

[hizzy7]

Hi

This is from a DVD, yes. One from Japan. Does it fall into the unfixable catagory? :P

Thanks!

[manono]

Depends on what you want to do with it. You could reinterlace it after unblending it as best you can. You could unblend it to 29.97fps as best you can (with blended frames during the sections that are greater than 29.97fps). You could just leave it alone. The Japanese don't usually mess up their DVDs the way the Chinese, the Indians, and some US companies do with European stuff. So whoever made this saw the problem. You'll notice only the action scenes are blended and interlaced while the slower scenes aren't. Is this from a TV show, or straight to video, do you know? I have no idea how it could be shown in a movie theater. Below is a link to one solution (progressive 29.97fps) using this script:

QTGMC()
CDeblend()
BlendBob()
LanczosResize(640,360)

If you want to unblend it to 59.94fps, QTGMC followed by CDeblend (SRestore won't unblend to the same framerate, I don't think), followed by reinterlacing it. But there's nothing really wrong with the source, I don't think. jagabo and/or others might have other ideas. It's an interesting source, that one.

[hizzy7]

He could unblend it to 59.94fps followed by reinterlacing it.

To make it 59.94, would I use QTGMC? What would you use to reinterlace?

Thanks!!!!!

[hizzy7]

Hi Manono,

Could you give me some tips on installing & loading cdblend? I created an avs file called cdblend, but it still won't load. I think I need the masktools filter and remove grain. Is there anything else?

Thanks!

h

[ndjamena]

c_deblend superseded by srestore

???

[manono]

Could you give me some tips on installing & loading cdblend? I created an avs file called cdblend, but it still won't load. I think I need the masktools filter and remove grain. Is there anything else?

So you found the function? The included doc only mentions needing MT_MaskTools.

[ndjamena]

Are you importing the cdeblend avs file? Otherwise you'll need to call it .avsi to load it automatically...

[hizzy7]

I have MT_Masktools-25 and MT_Masktools-26. Should I look for "MT_MaskTools" with no numbers preceding it?

Thanks!

[hizzy7]

Got it to work!

For the purposes of learning, why are we using LanczosResize(640,360)?

If I wanted to unblend it to 59.94fps, would I just use:

QTGMC()
CDeblend()

I would be curious to try that as a learning experience.

Thanks!!!!!!!

[jagabo]

SRestore() can deblend without decimating. See omode 1-5.

[manono]

SRestore() can deblend without decimating. See omode 1-5.

Oh, right, forgot about omodes. Thanks. The results should be very similar, with SRestore maybe giving slightly better unblending as compared to CDeblend.

For the purposes of learning, why are we using LanczosResize(640,360)?

Just to show an MP4 with a smaller size and in the right aspect ratio. For DVD, of course, you'd keep it at 720x480.

[hizzy7]

Hi!

Thank you for your help! I am wondering:

When using srestore, what omode would you recommend?

I have been playing with the script, just to learn a bit. I notice when I just use

QTGMC()
CDeblend()

I end up with a file that is 47.952 fps, not 59.94. How would I get a 59.94 file?

Also, does BlendBob reinterlace the video?

Thanks!!!

[manono]

When using srestore, what omode would you recommend?

I used CDeblend. Maybe jagabo or someone else can recommend one for your sample. Or maybe you can test the different omodes yourself, to see which is the best for your movie.

Also, does BlendBob reinterlace the video?

No, this is from the doc:

This filter takes a bobbed video stream and halves it's framerate by blending
every other frame with the adjacent frame that matches it best.

If no adjacent frame is identical, you get a single blended frame from the two (which happens sometimes during parts whose framerate is greater than 29.97fps), and if it finds a match it creates a single 'clear' frame. The result is always progressive. I'm only suggesting this as one possible solution. In your first post you called it a crappy video, but I think they did a pretty decent job with it.

[hizzy7]

Hi!

I am playing with the omode to see what works. 5 was pretty jerky. The video stuttered in places. I will try them all.

I would be interested in trying to unblend the video to 59.94 as a learning exercise. Would I use QTGMC() & CDeblend() without BlendBob() for this purpose?

Thanks to Manono & all

[jagabo]

Try this:

QTGMC()
SRestore(omode=1)
TDecimate(mode=1, cycle=20, cycler=9)

You can play around with cycle and cycler for the different sections.

To convert to interlace frames without blending:

ChangeFSP(60000,1001)
AssumeTFF() # or BFF
SeparateFields()
SelectEvery(4,0,3)
Weaev()

But, since you don't have 29.97 frames per second, playback will be jerky (on a 60 Hz display some frames will be seen for 1/60 second, some for 2/60 second). You might as well decimate to 29.97 fps, or just leave the video as it was, field blended.

[hizzy7]

Hi!

Would I use Tdecimate to decimate to decimate to 29.97?

Jerky playback is what I'm trying to avoid!!!!

Thanks

[jagabo]

Jerky playback is what I'm trying to avoid!!!!

You can't avoid jerky playback. Period. ~33 fps will always be jerky on a 60 Hz TV. 30 Hz is too low a frame rate for smooth playback of high action material. That's probably why they used field blending in the first place. I would just leave it that way.

[hizzy7]

In your first post you called it a crappy video, but I think they did a pretty decent job with it.

Hi manono!

I am curious - in what way did they do a good job? Should the fields not be blended?

btw, thank you for that clip you uploaded. I will make different versions with all of the scripts posted to see the differences.

[manono]

I am curious - in what way did they do a good job? Should the fields not be blended?

They wanted the real 29.97fps stuff to be progressive, which they did. It's the stuff that's faster than 29.97fps that presented the problem. They didn't have to blend those parts, but I think they decided to do so (blend some fields) to get the smoothest playback, and I think I agree with them. A case can be made for doing it that way. I spend a good part of my video encoding hobby life unblending crappy PAL2NTSC sources, but this is something different. You mentioned earlier you thought it was PAL to NTSC, but hope you realize now it's not. Then they could have done it the way jagabo demonstrated earlier - interlacing unblended fields and dropping some during the speeded up parts, also one of my earlier alternative suggestions.

Sometimes, because of the much higher bitrate interlacing needs for the same quality it's necessary, when possible, to make something all progressive (by IVTC or unblending or, in this case, purposely blending). That's what my sample upload demonstrates. Sometimes it's difficult to fit an interlaced movie onto a DVD5 at decent quality. That might also be a consideration here, depending on how much of this video is actually interlaced. For example, if only 10% of it or so is those speeded up fight scenes, I'd probably either keep it the way it is (blended fields) or do something similar to what jagabo showed before.

[jagabo]

I am curious - in what way did they do a good job? Should the fields not be blended?

The theory is that blending fields gives the appearance of smoother motion.

If you don't blend fields you simply repeat fields (or frames). When the new frame rate isn't an integer multiple of the old frame rate jerkiness will become apparent. For example, with 33 Hz to 60 Hz most frames will be displayed twice. But some will be displayed only once. That creates an visible jerks several times a second.

See the attached video where a 33Hz video has been converted to 60 Hz with and without blending. If you step through frame by frame the blended portion (bottom row) looks terrible. The non blended part (top row) is sharp but some frames are displayed twice, some only once. At full playback speed the duplicated row is jerky, the blended row looks much smoother -- the blending looks like motion blur.

[hizzy7]

Hi!

I have found this discussion quite illuminating. I see how blending fields can offer smooth playback, and how blended fields can be desirable at times. I find the blended fields less distracting than the jerky playback. Smoother motion looks good, and, if I am understanding this correctly there must be a compromise in regards to the blended fields so as to keep the action scenes looking fluid.

Thank you again for your help!!!!!

[ndjamena]

Try to keep in mind, the authors of these videos don't have the possibility of future re-encoding high on their list of priorities...

[2Bdecided]

I'd be amazed if the author of the video "thought" about this at all, beyond running the time-stretch function in their NLE, and accepting whatever output that gave with the project setting he/she was using.

The only way to change the frame rate like this without blending is to use motion compensated frame rate conversion, and that can introduce artefacts, especially with fast motion. It's great for preventing panning from stuttering, but not for action shots. IME. YMMV.

Cheers,
David.