Pal to NTSC conversion ;o)

[StrobeLightEpileptic]

I've read the sticky. I've been reading a lot and experimenting quite a bit in the last few days on the topic. I'm currently using this tutorial:

http://www.digitalfaq.com/guides/video/convert-pal-ntsc.htm

This is supposed to have been recently updated, but I have a couple of questions on it. (As far as the propriety of posting a question on this forum about another site's guide... well, I got the link for the guide on videohelp.)

The first question I have is on his choice of deinterlacing filter. He wants to use VirtualDub's "Deinterlace Area-Based" filter for de-interlacing. But on Doom9 it's suggested that a person use the DeinterlaceMAP plugin for VirtualDub. Does anyone know the difference between the two filters to say why one would be better than the other?

My second question though is my main question, and that concerns changing the frame rate from 25fps to 23.976 fps. I just do not understand this latter frame rate. To quote from the tutorial I'm using, he says:

"VirtualDub, however, is going to temporally shift frames. For example, one second of 25fps PAL video will become one new second of video with about 24 frames, leaving one extra frame for the next second. Our video will be about 4 percent longer, as every 24 extra frames becomes one more extra second. This will cause audio to lose sync, and that will be addressed in the next step. This shifting of frames is how quality is maintained, and how further ghosting/blurring is avoided.

"Trivia: If the ultimate source of this video was NTSC to begin with, before it was PAL, then all we're doing it restoring it to its original length. Believe it or not, most PAL videos are sped up. When needed, the audio pitch is also adjusted by 4 percent to compensate for the "chipmunk" pitch that can sometimes start to creep in. "

Well, he's right in that the PAL video, (this is an old PAL vhs), I'm converting was originally sourced from NTSC. But I'm at a loss to understand what is happening here.

Now, once he has put this through VirtualDub he sticks it into TMPGenc with two settings:

1) Frame Rate: 23.976 (internally 29.97)

2) Encode Mode: 3:2 pulldown when playback

Is this what corrects the video so that it becomes 29.97? Is this telling the dvd player to perform this pulldown on film footage? And any old dvd player can do this?

I can try this and find out for myself, but with TMPGenc I'm in for a long encoding time so I'd like to get it right, or at least close, the first time. I'll probably set this all up tomorrow afternoon.

[StrobeLightEpileptic]

Well, as a follow up. I have found the smart de-interlace filter to my liking.

But, I am having a deal of trouble with the 23.976 framerate. Following the guide, I first ran the video through the filters in virtualdub. I then time-warped the audio to match the time of the resultant video.

I then went ahead and encoded the avi with Tsunami using the settings I mentioned above.

I burned this to disc and found that the video was jumpy in scenes involving quick motion, but the video was also WAY ahead of the audio.

I've been going through various threads about this subject of Pal video conversion to NTSC but I just don't find a lot of discussion about this framerate conversion and internal pulldown.


Can anyone point me to a thread, discussion, tutorial where I might be able to investigate what went wrong with my result?

Thankyou.

[yoda313]

I do have a manual, realtime alternative for you. If you don't need the original menus and don't need 5.1 surround sound you can do a real time dub.

Just use a dvd player that plays pal and dub it to a dvd recorder or to a computer capture device. Than the synching to ntsc will be done automatically.

Yes you will lose menus and 5.1 surround but if all you really want is the video and stereo will suffice this method is a lot less frustrating than software methods.

I have done this only once but it worked for me. I used a soft modded xbox 1 that played the pal dvd without macrovision. I then copied it to the hauppauge pvr250 caputure card I had at the time (long since sold on ebay). It worked like a charm. I got a nice ntsc mpeg 2 file I was able to author to a normal ntsc dvd.

There are no discernable issues on the dvd copy.

Again this might not be the best alternative for you but if you just want to get a viewable copy this method might be best.

If you don't have an external dvd player that plays pal dvds you could play them on the pc and then record to either another pc or a dvd recorder.

[FulciLives]

To be clear you have a PAL VHS video that you want to convert to a NTSC DVD ... correct?

So my first question is how are you playing back the PAL VHS video and how did you get it to a digital form? Did you use a computer based capture card/device or a stand alone DVD recorder or what?

In other words I think I can help but I need more information from you first ...

- John "FulciLives" Coleman

[StrobeLightEpileptic]

Hello again. Yoda, I understand what you are saying, but I am trying to capture from a Pal vhs tape. I suppose I could make it easier on myself by just doing the manual conversion, but when that is done, you're left with ghosting issues because of the pull down. That's why I am attracted to the idea of taking the source back to a 23.976 framerate, because that way you have progressive frames. The way it was shot to begin with. Also, I can't see where you can do anything with the colors this way. The reds in a Pal conversion are way too bright.

Fulci. I am using a Universal VCR and have it set for PAL video. It's outputting through composite video. From the VCR I have it going through a DataVideo TBC 1000. and from there into a Canopus ADVC110. I capture using WinDV and have it set for type-2 AVI.

Then I use VirtualDub to connect the individual avi segments and edit. First I save the Audio only as PCM. Then I save the AVI with no audio. Direct Stream. Once I have the one video file I take it back into VirtualDub.

First I apply the Smart de-interlace filter(2.8 beta 1). I have it set for Frame-only differencing. Edge-directed interpolate.
Compare color channels (instead of luma). Motion map denoising. Motion threshold = 15. Scene change threshold = 100.

Then I apply the resize filter. I disable aspect ration, (have to to get 720 x 480), and then type in absolute pixels of 720 x 480. I choose the Filter mode of Lanczos3.

I then apply the Static Noise Reduction filter with a setting of 6. (This seems redundant since denoising is applied also in the Smart de-interlace. But I've also left the SNR out once because of that and was not as satisfied. So I put it back in.)

Lastly, I adjust the framerate, changing it to 23.976 fps.

I then save this as an AVI. Once it's done I load the result to get the time down to the thousandth of a second. Put the audio into Goldwave and do a timewarp to match the time of the AVI.

Now, I stick the AVI into Tmpgenc Plus 2.5:

In Environmental Setting I enable the Canopus DV Codec settings with the color space set for Basic YCbCr(Default) - I think this helps to tone down the flaring reds.

Video:
Stream type: MPEG-2 Video
Size: 720 x 480
Aspect Ration : 4:3 Display
Frame Rate: 23.976 fps (internally 29.97 fps)
Rate Control: 2-pass VBR(VBR)
VBV buffer 224
Profile: Main Profile & Main level(MP@ML)
Video Format NTSC
Encode Mode 3:2 pulldown when playback
YUV format 4:2:0
DC comp prec 10 bits

Motion search pre = Highest Quality(very slow)


Advanced Settings:
Video source type: Non-interlace(progressive)
Field Order Top Filed first(field A) -- I suspected this caused my jitter, but this was done auto by Tmpgenc.
Source AR 4:3 525 line (NTSC, 704x408)
Video Arrange Meth Full Screen

Again I applied Noise Reduction and Clip Frame.

GOP Structure: the default except I deselected Closed GOP

I output this as ES Video only. And author it with it's PCM file converted to 192kbps ac3

I have gone over your guide trying to convert this with VirtualDub, But there to I had a problem with the audio video sync. But I am unsure now about any video jitter.

I appreciate any advice you can give me on this. Thanks.

[FulciLives]

Would it be possible to upload a small section of the original DV.AVI file that you have captured?

Maybe cut out a short section of it with VirtualDub and save using DIRECT STREAM COPY so nothing is changed.

RapidShare.com allows for file upload and download ... all free ... just have to make sure the file is not larger than 200MB although it doesn't need to be that big per se ... I just need a few seconds or so and it would be nice if it is a scene with a more or less steady frame but with movement inside the frame (like someone walking from left to right or something like that).

- John "FulciLives" Coleman

*** EDIT ***
OK I just noticed that you change the frame rate in VirtualDub (was re-reading your process). So If I understand correctly ... before feeding it to your MPEG-2 encoder (in this case TMPGEnc Plus) you have a DV AVI file that has been processed so that it is 23.976fps ... right?

Have you tried playing that back? Maybe using VLC etc.

The point is to isolate the problem so in other words ... is the problem the MPEG-2 encoding OR is the problem what you are doing prior to that? If you can play back the processed DV AVI and it plays back smooth then you are doing something wrong in TMPGEnc which should be relatively easy to correct however I suspect the problem is in your VirtualDub processing which is why I'd still like to see an unprocessed sample (I don't care if it has audio or not). Also it could just be that some "oddness" is going on with the original source (the PAL VHS) i.e., maybe it was converted from NTSC as opposed to FILM etc.

So would still like a sample to "play" with

[StrobeLightEpileptic]

Would it be possible to upload a small section of the original DV.AVI file that you have captured?

Yeah, I have no problem with that. But I started another experiment last night that won't be done until tomorrow. So, I'll try to upload a section sometime tomorrow afternoon and get back.

*** EDIT ***
OK I just noticed that you change the frame rate in VirtualDub (was re-reading your process). So If I understand correctly ... before feeding it to your MPEG-2 encoder (in this case TMPGEnc Plus) you have a DV AVI file that has been processed so that it is 23.976fps ... right?

Have you tried playing that back? Maybe using VLC etc.

The point is to isolate the problem so in other words ... is the problem the MPEG-2 encoding OR is the problem what you are doing prior to that? If you can play back the processed DV AVI and it plays back smooth then you are doing something wrong in TMPGEnc which should be relatively easy to correct however I suspect the problem is in your VirtualDub processing which is why I'd still like to see an unprocessed sample (I don't care if it has audio or not). Also it could just be that some "oddness" is going on with the original source (the PAL VHS) i.e., maybe it was converted from NTSC as opposed to FILM etc.

So would still like a sample to "play" with

I edited my original post because I forgot to mention that I did set the Framerate in VirtualDub. So you probably read my post once before I made that edit. I also forgot to mention that when I saved all that filtering in VirtualDub I saved it using the huffyuv codec. Don't know if that matters.

But, yes, I see what you mean and tomorrow I will redo some of that so I can answer your questions and upload a segment.

thanks.

[FulciLives]

I'll do my best to help

[lordsmurf]

Just use a dvd player that plays pal and dub it to a dvd recorder or to a computer capture device. Than the synching to ntsc will be done automatically.

Low quality output -- not suggested.

[lordsmurf]

1) Frame Rate: 23.976 (internally 29.97)
2) Encode Mode: 3:2 pulldown when playback
Is this what corrects the video so that it becomes 29.97? Is this telling the dvd player to perform this pulldown on film footage? And any old dvd player can do this?

NTSC is 23.976 too, not just 29.97. Yes, any DVD-Video player can play this.
Most commercial DVD releases are this way.

[lordsmurf]

ghosting issues because of the pull down.

Pulldown doesn't cause ghosting.
Changing framerates poorly is what does it (yoda's method, for example).

But there to I had a problem with the audio video sync..

Using Goldwave to convert length?

I apply the Smart de-interlace filter(2.8 beta 1).

Not the suggested filter for best results.
Not everything your read at doom9 should be considered useful.

[StrobeLightEpileptic]

Okay. I've uploaded a 100MB clip. That's probably a little more than you needed. Sorry if it's too much. The link is here:

http://rapidshare.com/files/238014940/clip.avi.html

MD5: 21919D5BDBF1CF6E2D004E212C234D36


****EDIT****
Smaller clip is posted here:

http://rapidshare.com/files/238233812/smaller_clip.avi

**** END OF EDIT****

Last night's experiment ended suddenly when I tripped over the ac adapter wire cable of my external harddrive. Knocked the harddrive off the table and it would only beep from there on.

So, I got a new harddrive this morning. I was hoping I would be able to view the filtered avi from VD but now that will have to wait until this afternoon. I am processing that now, except this time it's uncompressed output instead of the huffyyuv.

I sure do appreciate any advice you can give, because I would really like to know how to do this correctly.


LordSmurf:

As I understand it, NTSC interlaced is 29.97 and I understand that. NTSC progressive is 23.976 ? Otherwise known as filmrate? See, the term filmrate with that number somewhat confuses me because true filmrate is commonly 24fps. But this being video I suppose the fractional quantity is because of the adjustment between B&W and Color?

Thanks for clearing that up about framerate causing ghosting instead of the pulldown. I thought it was pulldown because of the additional frames you had to add in. I once had a chance to use Canopus Procoder 3 for a PAL conversion to NTSC and it didn't remove the ghosting either. I've always been told that Procoder 3 was such a great piece of software for this and always did a good job. But, it would seem that it would at least get the framerate conversion correctly.

Yes, I did use Goldwave to convert the length of the audio.

You say the smart deinterlace filter is not the best, but you do not say why. Are you saying that the deinterlaced - area based filter suggested on digitalfaq is the one that should be used? Can you explain why, or is this just experienced based?

I do appreciate the input.

Thanks all. Will get back on how well the filtered AVI plays in VD later this afternoon.

[manono]

You're having us download a 100 MB AVI? That's how you capped this thing - as an AVI? And is 100 MB really the smallest size you know how to cut? Or is this DV AVI maybe?

NTSC progressive is 23.976 ?

29.97fps can also be progressive, although even if the source is progressive it's usually encoded as interlaced for DVD.

But this being video I suppose the fractional quantity is because of the adjustment between B&W and Color?

No, because NTSC requires 59.94Hz.

And that's a horrible guide, in my opinion.

[StrobeLightEpileptic]

You're having us download a 100 MB AVI? That's how you capped this thing - as an AVI? And is 100 MB really the smallest size you know how to cut? Or is this DV AVI maybe?

Yes, sorry. DV AVI type 2.

But this being video I suppose the fractional quantity is because of the adjustment between B&W and Color?

No, because NTSC requires 59.94Hz.

It was my understanding that before color ntsc video was an even 30fps. And the onset of color was why it was reduced to 29.97 fps. It was either that or create a new scheme that would not be backward compatible with B&W. I can't remember where I read that now.

And that's a horrible guide, in my opinion.

Hey, I'm always open to looking at a better guide. Suggestions?

Thanks.

[StrobeLightEpileptic]

And is 100 MB really the smallest size you know how to cut? Or is this DV AVI maybe?

Okay, sorry. I just wanted to make sure I provided enough video, but that does take a while to download. Here's a smaller clip:

http://rapidshare.com/files/238233812/smaller_clip.avi

30MB I hope that's enough. But if it's not you can go with the larger clip. I've also edited my above post to include this clip.

Thanks.

[manono]

It's badly converted from NTSC to PAL and is field-blended garbage. The 'true' framerate is already 23.976fps after being unblended and the duplicate fields/frames removed. Only AviSynth can do it properly. Here's the script I used:

AVISource("E:\Test\Clip.avi",false)
Yadif(Mode=1,Order=0)
SRestore()
Crop(12,4,-12,0)
Lanczosresize(512,384)

Of course, if for DVD you wouldn't resize, although you'd probably want to remove the uneven and ugly black on the 3 sides and put on fresh black pillarbars on 2 sides. I made a 4.5 MB 512x384 XviD of it that you can find here:

http://www.mediafire.com/?dmznfz2myr2

http://avisynth.org/mediawiki/Srestore

Edit: Loking at the AVI again, I now notice garbage along the bottom. Before resizing I'd crop:

Crop(14,4,-14,-4)

[lordsmurf]

The guide is fine. In fact, I know it is, it works great, resulting in minimal problems found by most other methods. The issue you've run into is that your source video is damaged, and this guide does not address such things.

Manono has given you an alternate method that will need to be used to de-crapify your source, attempt to restore it, remove the ghosting.

But I have a feeling that it will still look somewhat crappy in the end. Never good to hear, but needs to be said.

I tried to download the rapidshare clips, but the download kept screwing up at the end of file. (I hate rapidshare, wish people would stop using it.) What manono posted looked fine for motion, but there is block noise and color issues that really need addressing in VirtualDub, with more filter work.

[manono]

What manono posted looked fine for motion, but there is block noise and color issues that really need addressing in VirtualDub, with more filter work.

I purposely didn't try and improve on those aspects of the video, but only fix its greatest problem - that his source is field-blended.

And the guide is not fine, in my opinion. That someone in this day and age would present a guide going the VDub route all the way when the filters, encoding speed, results, are much better by frameserving via AviSynth, indicates this person's incompetence. That guide cannot address the main problem with that video because there's nothing the VDub filters can do to correct the problem. Any solution involving deinterlacing alone (which is the guide's answer to the problem) is crippled. That kind of NTSC to PAL conversion is not all that uncommon, and not some kind of an aberration.

In addition, there's so much wrong with that guide I don't even know where to begin. First, it talks about being a guide for converting interlaced PAL videos to NTSC. That's a fairly small percentage of the PAL videos out there. If it were a movie simply speeded up to 25fps, all that nonsense about deinterlacing would be superfluous and damaging to the video as it would be progressive already. If it really were a true PAL interlaced video (not NTSC2PAL), the chances are good - very good - that its true framerate is supposed to be 25fps, and neither the video nor the audio needs to be slowed for conversion to NTSC. And, as I said before, it can't properly address the kind of interlaced PAL video that StrobeLightEpileptic has, except to offer deinterlacing as a solution. And if it was NTSC converted to PAL by field blending, the length remains the same and neither the video framerate nor the audio needs slowing, another big error in that guide. As evidence you have that 23.976fps AVI I made which didn't slow down or lengthen the source video.

The issue you've run into is that your source video is damaged, and this guide does not address such things.

Aah, but it does, and very foolishly. Here's what the guide says:

Because of this, know that the PAL video you have already has artifacts, typically ghosting or blurring. So this guide does its best to minimize adding further artifacts. Sloppy methods DOUBLE the amount of ghosting and blurring.

And nothing is sloppier than just deinterlacing it using inferior VDub deinterlacers.

[lordsmurf]

AVISynth is needlessly complex, VirtualDub is a very useful GUI-based solution. Yes, among the download/warez crowd, interlaced PAL is definitely a minority -- but that's not what that site is about. VHS to DVD is a primary aspect to the video information, and your PAL VHS tapes will be interlaced sources.

I disagree that the deinterlace filter suggested is inferior or damaging in any way. I've seen it used in commercial video workflows.

I don't follow your argument against audio. That's anti-math. If you convert a 24fps to 25fps, by shifting the frames, you'll end up with an audio shift of about 4% to match the shifted video frames. There's no way around that.

In theory, the AVISynth filters used can be just as damaging. It really depends on the source video, to be completely honest.

There isn't a perfect way to convert, even studios acknowledge this one. Aside from Snell & Willcox hardware, both of the methods suggested here (the digitalFAQ.com guide, and the AVISynth method you've suggested) will work quite nicely. It just so happens that this source video may react better to your suggested method, as that one is specific to addressing the source errors on motion and blending.

Not aimed at you, but I get aggravated by the notion that AVISynth is superior -- it's not. It's just different. Sadly, it can be confusing and difficult to use, even for those experienced both by programming/coding and videography. And that's a shame.

[manono]

I don't follow your argument against audio. That's anti-math. If you convert a 24fps to 25fps, by shifting the frames, you'll end up with an audio shift of about 4% to match the shifted video frames. There's no way around that.

Look, they did the conversion by, in effect, adding one frame per second. Some such conversions really are done by adding one frame per second, or 2 fields per second. The framerate is higher but there are more frames. They balance out and the length remains the same. When 3:2 pulldown is applied to film to change 23.976fps to 29.97fps, duplicate fields are added at playback and the length remains the same, doesn't it? And the audio doesn't have to be converted, does it? The principle is the same here. After being unblended you have 25fps with a duplicate frame every second. That dupe frame is removed to make it 24fps (or 23.976fps). The length remains the same. That's the only good thing about that godawful conversion method - no PAL speedup and no off-pitch audio. The guide is 100% wrong.

I disagree that the deinterlace filter suggested is inferior or damaging in any way. I've seen it used in commercial video workflows.

What I said was that you don't deinterlace simply speeded up and already progressive 25fps movies. Surely you'd agree that deinterlacing unnecessarily is damaging? And you don't deinterlace field-blended NTSC2PAL converted videos when there's a better way. I defy you to follow that guide and then claim that deinterlacing that POS source that StrobeLightEpileptic has will produce results better than unblending it using AviSynth filters. And in cases where deinterlacing really is necessary, AviSynth has better ones - some of them much better. Actually, I find it surprising that you're backing the use of deinterlacers as I think you're one of the ones that advocates keeping things interlaced when for DVD. What I did wasn't, strictly speaking, a deinterlace. I unblended it and removed the dupe frames. It's more akin to an IVTC.

AVISynth is needlessly complex

Not once you learn it. Yes, it has a learning curve, but if you work with sources like that one, then you had damn well better learn it.

Yes, among the download/warez crowd, interlaced PAL is definitely a minority

What's that supposed to mean? That AVIs have to be deinterlaced? Even the encoders for the download/warez crowd know how to unblend field-blended video, well, some of them anyway.

In theory, the AVISynth filters used can be just as damaging. It really depends on the source video, to be completely honest.

That guide talks about exactly the kind of video that StrobeLightEpileptic has. VDub can't handle it properly; AviSynth can. There are thousands out there that could return that video to its proper progressive and unblended 23.976fps. The guide writer is one of many that has no clue how to do it properly.

Not aimed at you, but I get aggravated by the notion that AVISynth is superior -- it's not.

It is, in my opinion. Since you haven't used it, as far as I know, then I wonder how you're in any position to judge. Me, I started out using VDub and its filters, IVTCing with TMPGEnc when necessary, frameserving with a VFAPI. When AviSynth became robust enough to use (meaning mostly when it began to have IVTC filters), I jumped ship and haven't looked back.

[StrobeLightEpileptic]

Okay, everybody. My head is swirling. I have a lot to ask and speak about and I don't think I'm going to be able to this in an organized way. So, please read through the entire post before you respond and give me some leeway.

I do want to thank everyone for their input. Even if there are differences, I gain by the discussion.

So, where I left off was with Fulci. I told him what settings I used in VirtualDub when I ran my capture through it. He asked me about how it looked after I was done. I never looked at that. I just ran everything through Tmpgenc. So, tonight I first ran it through VD and compared it to the original capture in VD. It ran, but a lot slower than the original capture. About 1/3 the speed of the original capture. Running both at the same time might have slowed my system too, I don't know.

Then I ran it in VLC. It wouldn't run. I could drag the timeline to various scenes but that's it. Also the color was all wrong. I've posted a pic. The faces are blue. The original capture played fine in VLC.

I was able to play it in Media Player but it played even slower than in VD. The colors were correct though.

So, that's where that stands.

Now, both manono and lordsmurf have said the video was damaged. But you didn't tell me what you used to analyze it that you could come to that conclusion. I'm not disputing with you, but I'd like to know how to determine that myself. This isn't the only PAL video I want to work with.

And I really don't understand how it is damaged except you seem to be saying that it was 23.976 all along and that it was field-blended. So, are you saying that when I de-interlaced and resized it and then changed the framerate -- all of that just distorted it even more?

The biggest part of the problem I came in with was that my audio and video was out of sync, with the video waayyy ahead of the audio. manono seems to be saying that all of that is corrected by his avisynth script?

I'll be honest and say I was really hoping for a solution that avoided AviSynth. But, it does look like I'll have to do some studying up on AviSynth. I did try to replicate the script you suggested, manono. (Understand I've never played with AviSynth before), This is what I used:

loadcplugin("C:\Program Files\AviSynth 2.5\plugins\yadif.dll")
avis_clip = AVISource("C:\clip.avi", false, fourCC="CDVC")
yadif(avis_clip,1,0)
SRestore()
Crop(12,4,-12,0)
Lanczosresize(512, 384)

I had to use CDVC because the program complained about YUV for yadif. My result did look very much like yours.

And when I checked out "avis_clip" in GSpot it did show a framerate of 23.976.

So, if I run the entire capture through this script, you're saying I should have something I can put into TMPGENC? Do I need to adjust the audio in anyway?

Also, you say I should not even have to resize it. Are you saying it will come out to 720 x 480 in Tmpgenc?

I'm not assuming that the entire thing will be perfect, but I'll at least have something I can look at on the TV while I dig into AviSynth looking for other ideas.

Of course, if there is still a VDub solution, I'll listen, but I'm willing to face the fact that I might have to turn to AviSynth.

I think that's all I have to say. Not sure. I might have missed something. But, if so you can bet I'll be back ;o)

Thanks again.

vlcsnap-00001.png

[StrobeLightEpileptic]

Oh, yeah. Reading the recent posts, I realized another question about that AviSynth script:

The yadif arguments call for doubling the framerate. I've really tried to understand how you can do this and still have a framerate of 23.976.

I understand that doubling the framerate smooths out the video, But if you double the framerate how can you still end up with 23.976?

I know this probably sounds stupid. But that really does confuse me.

[manono]

So, are you saying that when I de-interlaced and resized it and then changed the framerate -- all of that just distorted it even more?

By deinterlacing you didn't achieve the optimum results. By then slowing it to 23.976fps you wound up with slower motion than intended and audio a semi-tone too low.

manono seems to be saying that all of that is corrected by his avisynth script?

Yes, you don't have to stretch the audio. You can use the original untouched audio, assuming it's AC3 audio. If MPA or PCM WAV audio you might want to convert it to AC3, but you don't have to 'timewarp' the audio as described in that guide.

Good for you that you got an AviSynth script running.

So, if I run the entire capture through this script, you're saying I should have something I can put into TMPGENC?

Yes, but if for TMPGEnc you should add as the last line in the script:

ConvertToRGB24(Matrix="PC.601")

I think, but am not positive, that's the right matrix. If jagabo sees this he'll correct me if necessary.

Also, you say I should not even have to resize it. Are you saying it will come out to 720 x 480 in Tmpgenc?

My mistake. I resized for the AVI sample, but forgot that you're converting to NTSC DVD. The correct crop/resize/addborders might be:

Crop(12,4,-12,-4)
Lanczosresize(704,480)
AddBorders(8,0,8,0)

But you didn't tell me what you used to analyze it that you could come to that conclusion.

When I first took a look it seemed to be a movie. Movies aren't interlaced. I then separated the fields and saw the blending, indicative of a crummy NTSC2PAL conversion. The rest was easy - put on the Yadif/Srestore combo.

The yadif arguments call for doubling the framerate. I've really tried to understand how you can do this and still have a framerate of 23.976.

Yes, Yadif is being used as a 'smart bobber' there. The thing with these field-blended videos is that one of the 2 fields making up the frame is blended and the other is "clean". The SRestore part of that combo then dumps any blended fields and uses the good field as the replacement. It also removes the duplicate frames after the unblending is done. The results won't be perfect. It might not remove all the blending. Because it's using a lot of bobbed fields, in places where there are fine lines you might see aliasing or 'line jitter'. That can mostly be fixed by using better bobbers, but the tradeoff is time - it'll take a very long time to complete the encoding. Yadif is often considered the best fast deinterlacer/bobber. With any luck the end result will be a smooth playing and "cleaner" video.

[lordsmurf]

Are you using HuffYUV for compression? VLC doesn't play some codecs very well. Re-open the video in a video editor, and it will be fine. VirtualDub, Adobe Premiere -- take your pick. So that problem may not really a problem.

Sorry, but I find manono's advice to be nothing short of arrogant and incoherent. You won't ever be able to make heads or tails of stuff like AVISynth unless you want to spend ungodly hours screwing with it -- and for some unknown reason, the only people who seem to have any skill using it absolutely suck at explaining it. All I can say to you is "best of luck". Also understand that it's not a magical piece of software that some of its fanboy followers want to make it out to be.

Again, the first guide works perfectly on a clean interlaced PAL signal -- I've done it many times. When you have ghosting in the source, the new NTSC version would retain the same ghosting -- that guide was not meant for restoring video, only converting clean PAL to NTSC. Unfortunately, you won't get the results you want, with this particular video, using that method.

[manono]

Sorry, but I find manono's advice to be nothing short of arrogant and incoherent. You won't ever be able to make heads or tails of stuff like AVISynth unless you want to spend ungodly hours screwing with it -- and for some unknown reason, the only people who seem to have any skill using it absolutely suck at explaining it. All I can say to you is "best of luck". Also understand that it's not a magical piece of software that some of its fanboy followers want to make it out to be.

Arrogant? Incoherent? He got the script going and he's not very familiar with AviSynth. I must have done something right - provided some good information, a good script guide, a good link - something. The only thing I've been arrogant about is my attitude towards those unwilling to try something better than than their current methods. And my attitude towards that useless and pointless guide.

Again, the first guide works perfectly on a clean interlaced PAL signal

Now, just think about that for a minute. What kinds of PAL sources have a "clean interlaced signal"? Not speeded up movies because they're not interlaced (or if they are they're just field shifted and just need the fields realigned, and not deinterlaced). Not the kind of NTSC2PAL material he has because it's not "clean". That just leaves native PAL interlaced 25fps video. And if native PAL, the chances are good it shouldn't be slowed to 23.976fps because that would mean making the movement too slow and the audio too low. Some people might deinterlace it anyway but, using AviSynth and only AviSynth, it could be converted to interlaced 29.97fps for encoding as NTSC. And therefore that guide doesn't apply to PAL sources with a "clean interlaced signal" either. What it's for it's wrong about. What it might be useful for it doesn't cover. The guide is quite clear that it's for PAL sources that started as NTSC, were badly converted to PAL and now need to be converted back to NTSC. It's written expressly for the kind of material that StrobeLightEpileptic has. It just happens to be dead wrong about nearly everything it says. That and it's recommending methods clearly inferior to the ones I suggested.

I have no beef with you, Lordsmurf. In fact, after that MPEG-2 encoding comparison thread we were both involved in together I gained a newfound appreciation for you because I know you disagreed with me on what I was saying but left me alone anyway.

[jagabo]

loadcplugin("C:\Program Files\AviSynth 2.5\plugins\yadif.dll")
avis_clip = AVISource("C:\clip.avi", false, fourCC="CDVC")...

I had to use CDVC because the program complained about YUV for yadif.

By forcing another fourcc you've changed what DV decoder is being used. I suspect you have Panasonic DV codec installed because it only produces RGB output which Yadif() will not like. If that's the case, get rid of Panasonic DV Codec. It has many problems. Use Cedocida instead.

Now, both manono and lordsmurf have said the video was damaged. But you didn't tell me what you used to analyze it that you could come to that conclusion.

An easy way to look at each field is to use the Bob Doubler filter in VirtualDub. This splits each frame into the individual fields (quickly filling the missing scan lines with data interpolated from the scan lines above and below, ie, a simple bob) and shows them to you one field at a time in the output window.

And I really don't understand how it is damaged...

If you use the Bob Doubler filter you will see that some fields are a weighted blending of the two fields around it. These blended fields look like double exposures:

Field before:


Blended field:


Field after:


This type of field blending is very common with the realtime frame rate conversion hardware that broadcasters use. The only way I know of eliminating these blended fields is with AviSynth, with Yadif().SRestore() being one of the best options currently. There are no VirtualDub or TMPGEnc filters that can do this.

Yes, but if for TMPGEnc you should add as the last line in the script:

ConvertToRGB24(Matrix="PC.601")

I think, but am not positive, that's the right matrix.

TMPGEnc can accept either PC.601 or rec.601 RGB sources. The default is rec.601 so ConvertToRGB() will suffice. If you want to use PC.601 you have to elect to use the "Output YUV data as Basic YCbCr not CCIR601" option in TMPGEnc. Note: I'm talking about TMPGEnc Plus here. I don't know if things have changed in their newer software.

[lordsmurf]

Not speeded up movies because they're not interlaced (or if they are they're just field shifted and just need the fields realigned, and not deinterlaced).

Yes, but you're still overlooking things. That's all I'm getting at. You have a case of tunnel-vision right now!

When you alter frame speed (and audio) -- which is the most common method of conversion, for filmed sources (be it movies, TV or otherwise), by studios -- from NTSC to PAL, which is clean, and then re-broadcast it on air/satellite in PAL country, and somebody records it to VHS or directly digital, you have a clean interlaced version recorded to tape/disc. Yes, you have a lot of awful conversions out there too -- but in my many years of dealing with PAL, that's a minority of cases. Again, the guide being used by the OP (from the first post) is strongly geared at PAL VHS to NTSC DVD, or interlace PAL DVD to progressive NTSC DVD -- it wasn't created for fixing mangled discs, XVID downloads, etc. For simple PAL-progressive to NTSC-progressive conversions, ample guides are online already. But you can't follow those when you have interlaced footage. And when the sources are fubar, as was the case in this post, it's not necessarily going to clean up the problems -- more advanced methods, such as the ones you suggested, are the proper course of action. But sadly, it's not an easy one -- AVISynth is too much like Scenarist or Maestro (complex authorware), customizing a mySQL database, or programming a website in ASP.NET.

I have no beef with you, Lordsmurf. In fact, after that MPEG-2 encoding comparison thread we were both involved in together I gained a newfound appreciation for you because I know you disagreed with me on what I was saying but left me alone anyway.

Okay, fair enough, I'll back off some. I don't have a beef with you either, and I do read a number of your posts. I just have to strongly disagree with you here. I think you're attacking another method due to your personal preference, not because the other method is bad. (Maybe an inferior method in this exact case in this post, but as we all know, video can be a case-by-case sort of world!) I didn't have a problem with AVISynth, or even your suggestions -- those were all quite good! I just didn't like how you poo-poo'd the other method, and then echoed that AVISynth-is-always-best stuff. (AVISynth vs VirtualDub is very much like Mac vs PC or Kirk vs Picard -- as you probably know, I usually take the middle ground, and can appreciate both).

We're good. 8)

[manono]

Not tunnel vision. If I seem overly focused it's because of that guide which is entirely wrong. This is my speciality. I deal with probably 3 or 4 field-blended DVDs a week. I know these things backwards and forwards. Didn't you read what jagabo said about these kinds of videos:

This type of field blending is very common with the realtime frame rate conversion hardware that broadcasters use. The only way I know of eliminating these blended fields is with AviSynth, with Yadif().SRestore() being one of the best options currently. There are no VirtualDub or TMPGEnc filters that can do this.

Now:

When you alter frame speed (and audio) -- which is the most common method of conversion, for filmed sources (be it movies, TV or otherwise), by studios -- from NTSC to PAL, which is clean, and then re-broadcast it on air/satellite in PAL country, and somebody records it to VHS or directly digital, you have a clean interlaced version recorded to tape/disc.

With these things (the kind StrobeLightEpileptic has and the kind for which the guide was written) the film's length and the audio aren't altered. Even if you did follow the guide's advice and take the easy and inferior way out by deinterlacing them, you still don't adjust the framerate or stretch the audio. The film remains the same length as its NTSC source after the conversion to PAL.

If you alter the speed, then they don't wind up interlaced unless, again, they're just field-shifted which is easy to fix without deinterlacing. jagabo showed field blending. Let me explain field shifting, in case you're not familiar with the term. It's not uncommon on PAL DVDs and maybe it's also seen with PAL TV captures, but I wouldn't know as I'm not in a PAL area and I don't cap.

If the capital letter is the top field and the small letter is the bottom field, ordinarily the fields are aligned like so:

A B C D E F
a b c d e f

When field shifted they're out of phase:

A B C D E F
b c d e f g

and appear interlaced. The worst thing you can do is deinterlace when a simple field realignment can be applied. This is done using AviSynth. I have no idea whether or not VDub can do it, but I suspect not. If that example is what you mean by "clean interlace" after a speed up to PAL's 25fps then, again, the guide's advice is faulty. Yes, you might want to slow it to its original 24fps but no, you don't want to deinterlace it, which will damage it unnecessarily.

There's also a way to convert to PAL 25fps from a film 24fps source which involves adding 2 duplicate fields every second, one every half second. It's not a bad way to make the conversion as it also avoids the PAL speedup. It will appear that half of the video is progressive and half interlaced as in that field-shift example above. To fix that you'd realign the fields and pull out that one duplicate frame every second. No slowdown should be done to either the audio or the video as it's a conversion method that keeps the film the same length. Maybe you've seen that kind before.

And when the sources are fubar, as was the case in this post...

It may be mangled, but that's what happens with this kind of conversion - the kind where they take the NTSC source and just run it through a standards conversion box as described by jagabo earlier. These aren't uncommon; you see them on DVD and apparently you see them even more often on PAL TV broadcasts. And it's the kind that guide addresses. But these aren't speeded up and they aren't "clean interlace", if I'm understanding your use of the term correctly.

[lordsmurf]

I understand you, but you're not understanding me.

If a movie is a million frames long, divide by 24 to get the NTSC, divide by 25 to get the PAL. You play the frames faster for PAL. The audio is offset about 4% in pitch. This is how most commercial DVDs are made, and how many progressive conversions are made from NTSC to PAL. The signal is then broadcast, and finds itself interlaced. I don't know of any broadcaster that converts on the fly. It's not telecined, it's interlaced -- you can't IVTC it.

Yes, you can use crappy methods too, which give you all kinds of hazy, blurry, blended crap. And that goes both ways, especially older work (70s, 80s, maybe some 90s). Sometimes this can be repaired, sometimes not. Replacing bad frames with adjacent frames is one method, but that can introduce judder.

Better methods include randomly adding in frames (false padding to 25fps), but that gives you the same problem as 3:2 pulldown does, with all the judder. This is what your method address -- removing judder from 24>25 (be it bad frames or just excess frames), but it will be added back through the use of 3:2 later. So all that smooth motion on the computer may not be all that smooth in a DVD player later, either! I'm not aware of this being used much, but with the way video is these days -- not a standard in sight -- I'm not surprised at all.

[manono]

If a movie is a million frames long, divide by 24 to get the NTSC, divide by 25 to get the PAL. You play the frames faster for PAL. The audio is offset about 4% in pitch. This is how most commercial DVDs are made, and how many progressive conversions are made from NTSC to PAL.

That's all very well and good. But that's not what that guide is about. I've already quoted from it once:

Because of this, know that the PAL video you have already has artifacts, typically ghosting or blurring. So this guide does its best to minimize adding further artifacts. Sloppy methods DOUBLE the amount of ghosting and blurring.

Once you get "ghosting or blurring", then it wasn't created using that much better NTSC2PAL conversion method you described. Your description is for 'clean' (to use your terminology) interlace. What he's describing is exactly what StrobeLightEpileptic has - 'unclean' interlacing. And that field blending method is one way to keep it very smooth playing and keep it the same original length, with no need to stretch the audio.

Also, I can't quite decide if what you're describing is broadcast video that's been encoded as interlaced but is from a progressive source, or is really interlaced when examining the individual frames. Of course it can be (and usually is) encoded as interlaced while the frames themselves don't show any interlacing. If it's just been speeded up and still shows interlacing/combing then, again, deinterlacing is exactly the wrong thing to do as it just has the phase shifted fields. Movies, by definition, are not interlaced. And they should never be simply deinterlaced. Although an IVTC, or an unblend, or a phase realignment, will remove interlacing as a by product of what they do, none of those processes use an original framerate deinterlacer to do the job (that's to differentiate from a double framerate bobber which unblending does use).

But that's OK; if we're talking at cross purposes then I'll retire and StrobeLightEpileptic can decide for himself how he wants to treat that video and other similar ones he comes across. I've already given him the best advice I know how to give. If he has further questions about exactly how to apply the unblender or something similar I'll be happy to jump back in. Otherwise I've about shot my wad on this one. It's been nice talking with you.