Interlaced video - I want it :-)

[Chibi Jasmin]

Originally posted by Yusaku
BTW: AFAIK matroxes are the only cards, that can hassle-free, out-of-the-box output fully interlaced DVDs to TV-Out. Not to mention that you can minimize the DVD player and continue to work on something else on the computer while playing files :)

Well, you know any MPEG2-DShow-Decoder that does not deinterlace or some way to stop Cyberlink or Intervideo ones from deinterlacing?

[Swan]

Originally posted by Chibi Jasmin


Well, you know any MPEG2-DShow-Decoder that does not deinterlace or some way to stop Cyberlink or Intervideo ones from deinterlacing?

Yes! Use Ligos MPEG decoding filters. Version 3.0 and 3.5 do not deinterlace. It's the ones I use.

Also, I just came to think that it may be an idea to rename the deinterlacing filters from Intervideo and Cyberlink to .bak. I'll try this myself. :-)
Graphedit reveals a few suspects... Sadly though, I suspect the deinterlacing function is built into the MPEG decoding filters, not a separate filter. :-(

/Swan

[Chibi Jasmin]

Where can I get the Ligos filters, please? I will happily try them...hope they will work in a chain with css-encrypted dvds...someone recommended the elecard filters a while ago, but mentioned they don't work with css-encr. dvds...and in fact, unregistering ivivideo.ax and registering the elecard filter didn't work...well, I will try the ligos ones then...

btw: the deinterlacing really seems to be build into the filter...but in case of intervideo it can be configured via registry to bob or weave...just not deactivated...

UPDATE: Well, on www.100fps.com weave is called doing nothing to the frame...not sure, if this is really true for intervideo, but you also said, it does not deinterlace further up in the discussion...if it works, it should leave you with the original interlaced frame and playback at 25 fps...I know, this is not what you want, but as far as my experiments go, outputting 25 fps truly interlaced video with right field order etc. leads to fluid motion on tv. When I been playing with tv captures, 25 fps video with two fields per frame (true interlace, where each field is different point in time) worked fine on tv. I personally see no need in outputting 50 fps field based video...

I will try the weave function of windvd again now...

[Zhnujm]

you can force both windvd and powerdvd to use the weave option. no deinterlacing is then done. thats what its made for - to recreate a full frame out of 2 fields, when the 2 fields came from a progressive source.
any deinterlacing would be stupid in such a case.

[Chibi Jasmin]

Yeah, just got that far, too...the WinDVD Weave option works fine for displaying my interlaced content (mostly extra stuff on dvds...) on tv...great!

Thanx...I won't disturb the ongoing discussion then anymore...but I am still curious what the point is in sending interlaced material as 50 fps field based to the tv...

My interlaced content here 25 fps frame based (with two fields different point in time -> 50 fps field based) works fine, when sent as 25 fps interlaced frames to the tv (using Matrox card btw).

I understand it like this. TV updates it's picture every 1/50 second...it get's a new frame every 2/50 (1/25) second, so it has exactly the time to draw both fields from every frame you sent it. And if field order etc. is correct in the frame based interlaced stream, everything is fine...at least it worked for me...when I been capturing from tv this was the way it worked fine...I verified this with very critical content (newstickers scrolling from right to left), where I could see every field order/interlacing etc. problem...

I'd say, just keep the video 25 fps frame based with two fields per frame and encode it with an interlace option like the one from xvid (treat it as fields and encode every second line, first odd then even or something like that).

I don't think there really is a hardware, also not the matrox card with your mpeg-2 content, that sends 50fps field based. I think it sends exactly what I say 25 fps with two fields per frame...because that is what you see on monitor and matrox outputs exactly that...

...well, still doesn't solve the problem...why don't your encodes work!? They should work...without any 50 fps tricks...sometimes you have to change field order or shift the image one row up or down on playback, but this shouldn't be necessary, if you just encode the working mpeg-2 without modifications...good luck in further research...I just wanted to share my experience...

[Swan]

Chibi Jasmin: Yeah, just got that far, too...the WinDVD Weave option works fine for displaying my interlaced content (mostly extra stuff on dvds...) on tv...great!

Thanks for the suggestion, Zhnujm, and for trying it, Chibi. I'll try this on my DVD's too.

Thanx...I won't disturb the ongoing discussion then anymore...but I am still curious what the point is in sending interlaced material as 50 fps field based to the tv...
Please, don't feel like your're disturbing. I appreciate your input.
I have not said I wanted to output 50 fps field, have I? :-)
I just wish to output interlaced Xvid, like I output my MPEG-2 files.
I do not see 50 fps field based (splitting each field into 2 frames, effectively reducing the resolution in half) as an option at all.

I'd say, just keep the video 25 fps frame based with two fields per frame and encode it with an interlace option like the one from xvid (treat it as fields and encode every second line, first odd then even or something like that).
I selected Interlaced in the XVid options and I did not add any deinterlacing filter or filter that changed the field order. The strange thing is why it didn't work. If the MPEG-2 file played interlaced on the TV, how come the interlaced Xvid file created from the MPEG-2 did not? I did not modify it, other than resize it to 480x576 and encode it to Xvid. It was clearly sent interlaced on the TV, yet the motion looked strange. It puzzles me!
Perhaps I did something wrong in the resizing process?
If anyone's got any suggestions on how to write an avs file for svcd resolution and a software for calculating cropping and AR, please let me know. Perhaps GordianKnot is not suited for it?


I don't think there really is a hardware, also not the matrox card with your mpeg-2 content, that sends 50fps field based. I think it sends exactly what I say 25 fps with two fields per frame...because that is

I think so too.

sometimes you have to change field order or shift the image one row up or down on playback, but this shouldn't be necessary, if you just encode the working mpeg-2 without
Please tell me how I do that on playback, Chibi?

/swan

[avih]

@swan:

can u try pls with avisynth:

clip.bob
or
clip.complementparity.bob
(and let us know which worked smooth)
on your xvid clip.
play the avs file with wmp or zoomplayer or whatever.

if you have a low cpu/resuolution ratio or your cpu is at 100% all the time, then you could try replacing the bob function(s) with bob(0.3,0.3,clip.height/2) which will resize vertically by half.

does it look smooth? (you can try this both with tv out and/or monitor, they should both be smooth).

also, if you have ffdshow installed, and one of the resizing options is enabled, then you're lost. resizing kills interlaced content (unless it's a interlace-aware-resizing, which i've seen only in specialized filters but afaik, NOT in ffdshow). also, if you have any other resize filter in your dshow filters chain (divxg400 or bicubicresize or others), that's prob what screwing you. (try with graphedit to see if there are any hidden filters in your chain).

cheers
avi

[Chibi Jasmin]

Originally posted by Swan

Thanks for the suggestion, Zhnujm, and for trying it, Chibi. I'll try this on my DVD's too.


Please, don't feel like your're disturbing. I appreciate your input.
I have not said I wanted to output 50 fps field, have I? :-)
I just wish to output interlaced Xvid, like I output my MPEG-2 files.
I do not see 50 fps field based (splitting each field into 2 frames, effectively reducing the resolution in half) as an option at all.


I selected Interlaced in the XVid options and I did not add any deinterlacing filter or filter that changed the field order. The strange thing is why it didn't work. If the MPEG-2 file played interlaced on the TV, how come the interlaced Xvid file created from the MPEG-2 did not? I did not modify it, other than resize it to 480x576 and encode it to Xvid. It was clearly sent interlaced on the TV, yet the motion looked strange. It puzzles me!
Perhaps I did something wrong in the resizing process?
If anyone's got any suggestions on how to write an avs file for svcd resolution and a software for calculating cropping and AR, please let me know. Perhaps GordianKnot is not suited for it?




I think so too.


Please tell me how I do that on playback, Chibi?

/swan

Well, I played with this a while ago, so I don't remember every step I took exactly...first, you have to get the field order right...there is a picture in the thread about incorrect field order...as your mpeg2-content works, I assume you have it right...let's take a next step...compare one frame from your encoded video with the original video...should be the same (except some quality loss) concerning the placement of interlaced lines...then play with some DShow based player? The trick is to get the output match the PAL TV exactly...normally this should be no problem with matrox cards, if the clip is 720x576 and card is configured for pal tv, if the mpeg2 worked...but maybe the player somehow screws something...the only thing I can suggest is to kick DivXG400 into the chain and try it's 'shift image by one row' option...this shifts the image by one field, this helped in some cases...I don't remember anymore at the moment, I'm afraid...sorry...

Another thing to try is to do the same thing, but just encode to DivX3 or 5 just for reference...these codecs don't have any special interlace option, but if you keep quality high enough, it should be suited for testing, if they screw up on playback also...

Hope this helps...good luck...

Ah yeah, and don't use DivXG400's overscan compensation option or some other video manipulating filter while testing...

[Owen]

Swan,

Did you say that your Xvid test file was made from an Mpeg2 Capture?
It would be a much better test to capture direct to Xvid.
I suggest 720-576 YUV2
Xvid set to 1pass Quant. Mode.
Set Quant. to 2
Select Interlace mode (just for fun)
Set motion search as high as you can without frame drops.
I use HIGH on a P4 2.1Gig WinXP
All other setting to default.

For capture I use FlyDS
http://asvzzz.narod.ru/

or Virtual VCR (freeware)
http://www.digtv.ws/

Hopefully one of the above will let you capture YUV2.
They both work for me.

These setting gives a data rate of around 3.5-4Gig per hour with uncompressed audio.

How dose that compare to your Mpeg2 captures ?

Quality is as good as original, limitted only by BT8x8 cature card with 8bit ADCs.

Cant wait for the new Conexant 10bit TV cards to become available.


I will reinstall PowerVCR and test.
This is going to be interesting.


Regards,

Owen.

P.S. If I capture in RGB24 I get a very dark picture and have to adjust capture driver settings to compensate.

[Swan]

Now, here are the results of my tests.
This was the procedure I used this time:
1. Create d2v project from interlaced MPEG-2 file (George Michael music video, 720x576 in size)
2. Create avs script, by hand, NOT using Gordian Knot.
3. The script contained two lines:
LoadPlugin("c:\program files\gordianknot\mpeg2dec.dll")
MPEG2Source("george.d2v")
Note that I did not use any arguments to crop or resize.
This clip played interlaced, perfectly smooth on the TV, using Media Player 6.4 and had the correct 4:3 AR! I used the Xvid playback filters (not ffdshow) and I did not load any additional filters. I did not see a green line this time (if there was one, it was in the overscan area, which doesn't matter anyway). :-)

I suspected the cropping I did in Gordian Knot to remove the black bars was the suspect. When I capture I get black areas around the bottom and left hand side in the clips (my card captures more than the overscan area). When encoding to DivX, I always remove these black areas, as using up bits on them is a waste. But, this is a no-no when encoding to interlaced Xvid! I wonder if you guys crop out the black on your captures that you Xvid-encode? If so, how do you do it and make sure you maintain the aspect ratio? If you've encoded to DivX using GordianKnot you know what a great resize/crop tool it is.

My next test was taking the same d2v file and changing the avs script to:
LoadPlugin("c:\program files\gordianknot\mpeg2dec.dll")
MPEG2Source("george.d2v")
BicubicResize(480,576,0,0.75)
This clip looked stretched (of course) when output with Media Player, so I had to use Bsplayer and set it to 4:3 for the image to be scaled correctly on the TV. This clip too, played perfectly and was interlaced!

So, there was never anything wrong with the Field Order. It was wrong to resize and crop the clip using GordianKnot! Or perhaps I used it the wrong way.
This brings me to an interesting conclusion: it's not possible to crop clips that are to be 576 in vertical resolution. It's "stick with the black junk or else".:-)
I'd love to know how you resize your 704x576 and 720x576 clips to 480x576, and if you crop off the black. I understand that cropping in the vertical area is bad (since it has the 576 lines, which is what is needed to get PAL interlaced content), but it should be OK to crop black junk out off the horizontal area. And since cropping in the vertical 576 lines is out of the question, this means I'm forced to waste bits on black junk (there's always black junk in the bottom area of the stuff I capture). Hmmmm.. :-)

Owen, I'll give Xvid capturing a shot. Thanks for the suggestions on how to set it up.

How dose that compare to your Mpeg2 captures ?
It's close. It depends on the settings I use. But at least 4 gigs per hour, using p and b frames and a video rate of about 6000 should produce 4 gigs. So the figure sounds reasonable.

Hopefully one of the above will let you capture YUV2
Yes, it will. All other software's I've tried can access YUY2 and all possible other color spaces *except* Ulead VideoStudio! I wonder where the h**l it reads the available color spaces from! I've mailed Ulead, but am not too hopeful they'll give me an answer I can use.

I will reinstall PowerVCR and test.
Try VideoStudio or DVD Workshop instead. Set it up not to deinterlace. Then you'll get the "crème de la crème", I promise. :-)

If I capture in RGB24 I get a very dark picture and have to adjust capture driver settings to compensate.
I get thin black and white lines over the picture, that come and go. :-(

/Swan

[Chibi Jasmin]

As you found out the hard way you have to keep the original captured 576 lines...anything else will screw the TV output...you can resize horizontally to everything you want...just keep the height 576! That's the trick...the reason is your tv displays exactl 576 lines, that have to fit exactly the interlacing in your video file. If you reduce the height of the video, the image is upscaled by your video card and is sent with 576 lines to the tv, but then the interlacing is screwed...this also means you have to keep the black junk at the top and bottom, but it should be okay to crop it at the left and right...after that you can resize to any XXXx576, depends on if you want the clip encoded with correct aspect ratio or not (setting it later in the player).

[Swan]

As you found out the hard way you have to keep the original captured 576 lines...anything else will screw the TV output...you can resize horizontally to everything you want...just keep the height 576!
Yes, and I feel like an idiot, Chibi! I should have known better. I better crawl under a rock and hide for a while. :-D
But I learned something valuable and I hope others who read this thread and may not have dared to ask the things I did, also had some use of this discussion.

Owen: It would be a much better test to capture direct to Xvid.
I tried the settings you gave me (using a res. of 720x576) but it didn't work for me. Massive frame dropping! :-(
I then tried lowering to 704x576, same result. At 352x288, there were no drops. I played around with the motion search setting and only at "low" could I capture for a while without frame drops. Capturing with Xvid compression must put much more of a strain on the machine than MPEG-2 because I never get any frame drops at video bitrates of 8000-10.000, highest possible audio quality (real-time layer II encoding), and I can encode using I-frames only. I also tried 1 pass Quality, enabling and disabling Interlace and changing the Quantization type. I'll try some more, but it didn't look promising. :-(

/Swan

[Chibi Jasmin]

Originally posted by Swan

Yes, and I feel like an idiot, Chibi! I should have known better. I better crawl under a rock and hide for a while. :-D
But I learned something valuable and I hope others who read this thread and may not have dared to ask the things I did, also had some use of this discussion.

:D Well, now you know it, have fun experimenting with capturing...maybe don't compress the audio in realtime and then capture to xvid...or just try the old way...capture high-bitrate mpeg-2 (or maybe huffyuv?)...and then compress to xvid...shouldn't be that bad...

One other thing I thought of, but it's more theory for now...

One possibility to cut the black junk at the top and bottom is, to find a video filter to add into the DShow chain, that adds black bars at the top and bottom on playback to make exactly 576 lines...but I don't know of any filter at the moment that could do it... :D DivXG400 does something similar at left and right, but that's of no use here...

BTW: I also got something out of this thread...the fact, the weave option from intervideo decoder is quite useful :)

One more thing...you said, the black bars at top and bottom are in your captured material? Just a tip...if they are not really black, but kinda noisy or something from the capture, you can try Crop and then AddBorders in your AVS Script to make them really black, this should save some bits when encoding...

[Swan]

realtime and then capture to xvid...or just try the old way...capture high-bitrate mpeg-2 (or maybe huffyuv?)...and then compress to xvid...shouldn't be that bad...
:D I did not use any compression on the audio in my Xvid capturing tests. Perhaps the 'puter isn't fast enough, he, he. But you bet I'll keep on experimenting. ;)When I can get back to MPEG-2 interlaced capturing with VideoStudio, all will be well again. I am quite bothered that I can't do it at the moment.

DShow chain, that adds black bars at the top and bottom on playback to make exactly 576 lines...but I
Yeah! That would be great! Hopefully someone with programming knowledge will have the need for it and construct one. :)

One more thing...you said, the black bars at top and bottom are in your captured material? Just a tip...if they are not really black, but kinda noisy or something from the capture, you can try Crop and
Thanks for the tip. No, the bars are pitch black. All three (!) of my capture cards capture more than the overscan area, leading to black stuff that I crop out when DivX-encoding. Perhaps all capture cards do this? How many pixels of black and where they are located (top, bottom, left, right) depends on the channel I'm capturing from.

Now I need to find a way to compress to Xvid (480x576) using GordianKnot if possible, for a compressability test (that's one of my favorite features of it, plus the crop and resize tool). I'll take this question to the GordianKnot forum. :-)

Again, thanks to everyone who helped me reach a conclusion as to why I was failing.

/Swan

[Owen]

Swan,
Sorry to here that your Xvid capture was unsuccessful.
You do need lots of CPU power.
I upgraded from Athlon 1.4 to P4 2.1 just so I could real-time capture to Xvid.

I have installed Video Studio 6.5 and cant get capture quality that looks as clear and natural as Xvid.
Mpeg2 suffers from noise/grain and pixel crawl in motion areas which I find distracting.

I've tried different setting, all with Performance set to 15.

Could you post some recommended settings?

I cant get YUV2 working in Ulead either.
RGB24 just does not look as good no matter how much I tinker with settings.

Regards,

Owen.

[norxh]

wow i just tried the xvid interlace option and the dffshow filter.

it seems to work great. i captured directly to 480x480 1500kbps from satellite, and the quality was near perfect, when displayed on the tv. i set the tv output to 640x480, and i used bsplayer to scale it properly. when playing, on the monitor you see all the ugly combing, but on the tv (27" jvc iart) it looks beautifal, even when paused there is no combing.

if i want to view on my computer, i just fire up bsplayer, using the latest dffshow build, check xvid, decode using xvid, and deinterlace (median works nicely for me), and again it looks beautifal.

so does anyone see anything wrong with this method for archiving video? by this method i mean, storing the video interlaced using xvids interlace option, and deinterlaceing as neccesary? are there any disadvantages? it should compress just as well correct?

edit: did some more testing, it appears that interlaced video with interlace checked, has far more artifacts, than deinterlaced video with it unchecked. this is dissapointing, since its supposed to be doing field based encoding, as -h said "It feeds lines to the DCT function in field-order, instead of frame-order. This results in much better compression (i.e. because DCT doesn't have to deal with scanline artifacts) without any quality loss, if the same quantizer is used. "

[midiguy]

swan, are you resizing vertically? that could really mess up interlaced content. don't resize vertically.

[trbarry]

Or if you absolutely must resize interlaced video before encoding, use the InterlacedResize option of SimpleResize. It will avoid blending the fields, much as Xvid does with the IDCT.

- Tom

[-h]

edit: did some more testing, it appears that interlaced video with interlace checked, has far more artifacts, than deinterlaced video with it unchecked. this is dissapointing, since its supposed to be doing field based encoding, as -h said "It feeds lines to the DCT function in field-order, instead of frame-order. This results in much better compression (i.e. because DCT doesn't have to deal with scanline artifacts) without any quality loss, if the same quantizer is used. "

By artifacts, do you just mean blocking? If your content is really interlaced, enabling interlaced coding should *never*, under any circumstances, look worse than progressive coding.

-h

[norxh]

well, it maybe just me but it looks like the interlaced video, with interlaced checked has alot more artifacts (blocking) than the deinterlaced video stored progressively. please tell me if i'm wrong.

here are 3 short samples from spongebob squarepants. i captured to huffyuv 480x480 in vdub. all are encoded to xvid 1000kbit cbr.

first (http://home.attbi.com/~thrackensditch2/testpredeinter.avi) , deinterlaced with avisynth, encoded to xvid, interlace unchecked.

second (http://home.attbi.com/~thrackensditch2/testwithinter.avi), encoded to xvid, with interlace checked.

third (http://home.attbi.com/~thrackensditch2/testwithinternocheck.avi), encoded to xvid, with interlace unchecked.

[avih]

Originally posted by norxh
well, it maybe just me but it looks like the interlaced video, with interlaced checked has alot more artifacts (blocking) than the deinterlaced video stored progressively. please tell me if i'm wrong.


if you deinterlace but keep the framerate (i.e. keep it at 25 or 30) then you have less (half) temporal information, since it's 25 images/sec instead of the original 50 images/sec when not de-interlaced. in this case, for the same bitrate, the interlaced method might have more blocks since it has twice temporal info. if you do de-interlace, the frame size doesn't change (i.e. still 480x480) but removing the interlacing artefacts makes the video more compressible, so less blocks, in the cost of half temporal resolution.

if your want to make a true comparision, you should compare the encoding with interlaced enabled against "separateFields.<whatever deinterlacer u use that's working on fields>" either at 480x480 or 480x240.

[Swan]

Wow, lots of replies. I have been away for a few days and thought this was a closed subject. I'm glad you're all interested in this in keeping the discussion going, though! :)

You do need lots of CPU power.
I have an AMD Athlon XP 1900+. Perhaps it's not fast enough?

I have installed Video Studio 6.5 and cant get capture quality that looks as clear and natural as Xvid.
Okay. Perhaps it's a matter of taste. ;)
But, just to make sure: Did you install the patch from Ulead? It's crucial, as it replaces the Ligos encoding filters with Ulead's MPEG Now! filters, which IMHO are much better!
I have noticed that among the most demanding things to MPEG-2 encode (scenes that will contain visible blocks) are scenes with smoke, fog and fire. But raising the video bitrate and fiddling with the GOP settings normally gives good results, even on fire, fog and smoke in my experiements.

I've tried different setting, all with Performance set to 15 Holy smokes! :) If I go above 8-9, I start getting dropped frames. I usually set the audio bitrate to 384, video bitrate to 8000 and under Advanced you could try setting the software to just encode I-frames. I can't access to my computer right now and check my settings (it's busy capturing "classic videos weekend" on VH1:)), but if there are any other options that I use that are important, I'll add them to this message later.

I cant get YUV2 working in Ulead either.
Really? Ohh!! And you're using btwincap drivers? Please, please, write to Eduardo who makes the btwincap drivers and tell him that you also have this problem. It works for him with Media Studio 6.5, he wrote to me, so if someone else mentions that it's not working, perhaps he can find it in his heart to look into it. Yes, I am really depressed about not being able to capture interlaced MPEG-2 with Ulead's software's (I prefer VideoStudio, btw). And I'm an Mpeg-2 nut, I know. :-)

midiguy: swan, are you resizing vertically? that could really mess up interlaced content. don't resize vertically.
I accidentally did. But no more! :) I learned the hard way. I should have known better, I must have left my brain on the bus that day I encoded my first Xvid test file.

/Swan

[midiguy]

Originally posted by Swan

I accidentally did. But no more! :) I learned the hard way. I should have known better, I must have left my brain on the bus that day I encoded my first Xvid test file.

/Swan

actually, something that I noticed is that you can resize vertically if it is resized to exactly half of the vertical resolution

example: resizing 640 x 480 to 320 x 240
not sure exactly how that works, I suppose its because since it is half, it could just discard one of the fields? or what it probably does is just blend the 2 fields exactly together... since it is exactly half, I guess that makes sense (the video looks de-interlaced BTW, as if you de-interlaced it by blending, just, you don't really get the annoying bluring or ghosting). any comments?

[sadownik]

To Swan:
Not only Ligos filters plays MPEG2 interlaced, Intervideo DVD XP pack play, BUT...only at integrated XP driver (nvidia).
When detonators installed (no matter which vers.) mpeg2 plays non-interlaced (!).
I suppose that might be with other players...
Next thing:
Divx5 and interlaced: Divx5.00 encoding videos are interlaced, BUT... divx5.01 and 5.02 no matter which option switch, output is non-interlaced:(.

dyrektor@hotmail.com

[mf]

Ok, after a few replies I started page-downing, cause I had an idea and I didn't want to read everything until posting. So if the problem is already solved, just bash me on the head. If your TV-out has a "flicker-filter", it might help to turn it off. That should display any interlaced video as truly interlaced. Even if you just drew it in MSPaint (I did that once, flicker hell! :D).

[Karl Beem]

DivX 5.0.4 supports interlaced content, encoding takes 50% longer. Currently, if you write the MV log, you get crap - otherwise, it looks fine