Anyone done IVTC to a Progressive Source? (Experts please!)

[homerpez]

I have a dilema.

I've been capturing with my ATI AIW Radeon board, 640x480, MPEG-2, 5.35MB/sec rate. I have "Deinterlace" and "Visual Masking" checked, and the output in DVD2AVI says "Progressive" when I run the MPEG-2 video through it as a test.

What I'd like to know is... how to make the 29.97 source video 23.976 (24fps FILM rate), so I can encode it more efficiently? ...or, a better way to word it... how do I perform IVTC on ANY "Progressive" video, to drop the frame rate yet keep all the "important" frames???


I know what you experts will say... why not just encode at 29.97 NTSC? All frames? There is a method to my madness...
I want to improve the quality of my SVCD encode by 20% (it was full-screen video, and at 29.97 it looks good, but not great)... I figure that by doing FILM rate, I will be ble to increase the quality per frame.

I also know what the experts will say to this... I can use any "solution" you offer for IVTC if it is in TMPGEnc, but I will then be encoding the final movie with CCE afterwards (don't ask)...

Can anyone help?

<font size=-1>[ This Message was edited by: homerpez on 2001-09-09 21:33:03 ]</font>

[KoalaBear]

Funny you should ask.

I posted a message on this topic awhile back: see IVTC, UVTC, we all VTC for IVTC for the details.

Based on the coding method you described, I would add these notes --

(1) For a 640x480 capture, change the last line of the AviSynth script to read: BilinearResize(480,480)

(2) I'm not sure MPEG captures will work properly because of the way it handles fields. You will probably have to make use of a substantially raw capture format like YUV9 or HUFYUV-compressed YUY2 to get the best results.

(3) DO NOT deinterlace prior to processing. IVTC requires both fields to reconstruct progressive frames in the correct sequence.

(4) You can frameserve the progressive video to CCE if you don't want to deal with a temporary file, but there is absolutely no reason to run the video through TMPGenc.

IVTC can deliver spectacular quality improvements when done properly. It's not a terribly difficult process to understand or to work with, but you have to play the game by its rules or it will amount to a colossal waste of time.

Good luck.

<font size=-1>[ This Message was edited by: KoalaBear on 2001-09-09 23:31:35 ]</font>

[homerpez]

ITVC IS HARD, when there's no clear pattern!

I tried this, and there literally is no pattern to it at all. Going solely by the frames that looked interlaced, it was wither no interlace for 20-30 frames, or alternately every other frame was interlaced... When done automatically, it always chose the wrong frames to drop, when done manually it went way out of synch!

AVI capture is not an option on my system. It would tax everything too great (only MPEG-2 works at this res, bitrate). Or I maybe could do 352x240 captures, but that hardly makes sense when I'm aiming for 480x480 SVCD!

Maybe not doing "deinterlace" during capture would help, but I tried before... and DVD2AVI still says "Progressive"...! Back to square one?

(P.S. - There IS a reason I run it through TMPGEnc, and it IS frameserving! It contains a GUI to resize, plus I can do things like accurate source range, framerate, audio delay, separate audio encoding relative to the frameserve, etc. that CCE cannot do. And no AVISYNTH!
Note in your post thread from before, same thing VidGuy is doing, but my souce is, sadly, progressive)

So, any other takers? Is there some process where I can skip ITVC and just drop frame rate?



<font size=-1>[ This Message was edited by: homerpez on 2001-09-10 00:29:44 ]</font>

[Beavis]

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>
On 2001-09-10 00:27:10, homerpez wrote:

Maybe not doing "deinterlace" during capture would help, but I tried before... and DVD2AVI still says "Progressive"...! Back to square one?

</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

It's not a matter of helping or not, if you're deinterlacing you're losing the information needed to inverse telecine. Period. KoalaBear has given you a good answer

By the way, even though you said not to, why are you encoding in RealTime and then re-encoding with CCE?

[homerpez]

MPEG-2 640x480 5.32MB SOURCE (For high-quality, no noise)

MPEG-2 480x480 VBR 500, 1170, 2500 SVCD (Quality source matters)

Basically lowing the bitrate to 1170, 1800, or anything else with MPEG-2 capture, on-the-fly, produces not NEARLY the result I get with re-encoding from the high-res source.

This is a bit new, I admit, I'm used to doing forced FILM thru CCE, and having it look phenomenal on my TV... first time I tried this with anything but TMPGEnc...

Curious, now... does "Visual Masking" happen to mean "make progressive" on the MMC 7.1 dialogue? Or is there no way to capture interlaced?

FYI: to answer your question further, I'm not encoding twice. I only use TMPGEnc to set resolution, resize, and source range. Then I save the "Video Only" project file (.tpr), and convert to an VFAPI AVI file, then CCE encodes it. Apparently things like IVTC filters work also. It is WAY easier than AVS files.

<font size=-1>[ This Message was edited by: homerpez on 2001-09-10 01:12:43 ]</font>

[kinneera]

When you capture at high res without deinterlacing enabled, do you see interlacing artifacts when you play it back on your computer? DVD2AVI is intended to work with DVD VOBs, so there's a possibility it simply isn't detecting the interlacing of your MPEGs properly.

What you are attempting to do is dangerous, in a sense. IVTC is really intended to restore video that was originally filmed at 23.976 fps back to that framerate. Video that was actually filmed at 29.976 fps isn't really supposed to be a candidate for IVTC - you will actually be deleting a part of your video, and probably shortening it. This may cause huge a/v sync problems even if you're successful.

With those caveats out of the way, it seems that if your source video is progressive, don't try to use IVTC at all. Just set the output framerate in TMPGEnc to 23.976. Just a thought...never tried it myself.

[Beavis]

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>
On 2001-09-10 01:08:10, homerpez wrote:
MPEG-2 640x480 5.32MB SOURCE (For high-quality, no noise)

MPEG-2 480x480 VBR 500, 1170, 2500 SVCD (Quality source matters)

FYI: to answer your question further, I'm not encoding twice.

</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

You ARE encoding twice- First is the high bitrate during capture and the second is afterwards to get your lower rate. However, as you said, you like this result better than doing the desired bitrate with the initial capture. The best result would probably come from capturing at 352X480 (assuming NTSC) using HuffyUV or an MJPEG set at higher quality levels. It would also make your processing more straightforward. It doesn't take that powerful of a system to do this.

[homerpez]

Unfortunately, 352x480 capture doesn't help me much either, since my target is the NTSC standard of 480x480, so I'd actually be losing some here too...


Now responding to kinneera:

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>With those caveats out of the way, it seems that if your source video is progressive, don't try to use IVTC at all. Just set the output framerate in TMPGEnc to 23.976. Just a thought...never tried it myself.</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

I think I had tried this once (before I even knew about IVTC at all), and I just did simple Non-Interlace/Progressive 29.97 to Non-interlace/Progressive 23.976 rate. If I remember correctly, it produced a correct-looking video, but some frames appeared missing (just didn't look "quite right"... since apparently it is meant to show all 29.97 frames)...

Now, back to what's bugging me about this...

I tried a capture again, using NO Visual masking, and NO deinterlacing. The end product did say "Progressive" again in DVD2AVI, but this time I could clearly see interlace lines in both DVD2AVI, and TMPGEnc.

So, I went ahead and did another "auto" IVTC using TMPGEnc, and came across another question... and it's dumb.

What do the "Red" colored frames signify in this window? Do they mean frames that will be encoded to the final product? or, do they mean frames what will be removed (because of interlace)? It's the one thing they don't explain in TMPGEnc in a pop-up dialog...

At least at first glance, it was taking interlaced frames and leaving them blank, while flagging non-interlaced ones in Red. But I would like to know if it is working right...

I'm also resigned to the fact maybe the TMPGEnc --> CCE method doesn't translate well for IVTC, I might have to just do IVTC encodes with TMPGEnc, and forced FILM with CCE (I'd like to use CCE all the time, since the MPEG-2 is much less blocky overall)

[Beavis]

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>
On 2001-09-10 12:55:17, homerpez wrote:
Unfortunately, 352x480 capture doesn't help me much either, since my target is the NTSC standard of 480x480, so I'd actually be losing some here too...

</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

If you can tell the difference between a 352x480 AVI capture converted to a 480x480 SVCD MPEG and a 480x480 AVI capture converted to the same sized MPEG I'd be surprised.

[kinneera]

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>
I think I had tried this once (before I even knew about IVTC at all), and I just did simple Non-Interlace/Progressive 29.97 to Non-interlace/Progressive 23.976 rate. If I remember correctly, it produced a correct-looking video, but some frames appeared missing (just didn't look "quite right"... since apparently it is meant to show all 29.97 frames)...
</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

What type of video are you trying to convert? If it's not a movie that had a film source, then what you describe is a fundamental problem with your process. Video recorded with a typical camcorder, off of a TV feed, or whatever, is going to be at a framerate of 29.97, so any method you use to reduce it to 23.97 is going to cause the deletion of some frames, IVTC included. It will never look quite right compared to the original source.

Now if the video is from a film source, then it presumably has been telecined. There was already a good discussion of this in another thread, but I will repeat it here for convenience. Telecining works something like this:

There are 4 frames in the film source:
A B C D

To telecine, double all of the and split into interlaced fields:
AA BB CC DD

Now add an extra field to frames B and C:
AA BBB CCC DD

Redistribute back to two-field frames:
AA BB BC CC DD

Now you have five interlaced frames (29.97 NTSC).

IVTC (should) simply remove the hybrid interlaced frame (BC), merge the interlaced fields back into progressive, and thus restore the original movie. This ONLY works correctly if it was 23.97 fps to begin with. If your video was recorded at 29.97 fps, it will look like AA BB CC DD EE - in which case it's hard to say what IVTC will do with it (most likely delete the CC, so you still have a total frame lost).

<font size=-1>[ This Message was edited by: kinneera on 2001-09-10 15:54:38 ]</font>

[homerpez]

I think maybe ITVC will work for me, since there does appear to be a ton of frames that do show this interlace effect (sometimes badly) between certain fields, and it seems to be definite interlace (IVTC is needed)...

But what I really need an answer to (for my ignorant question) is... what does the Red highlight mean on the frames? (IVTC User Setting filter, TMPGEnc)... does it mean only the Red will be used? Or only the Red are blocked out?

Then when I'm done, would I want "interlaced" output? Or "Non-Interlace"? (First "Settings" page, TMPGEnc, for output res, rate, and type)

Thanks!

[kinneera]

I'll try again: interlacing (and associated artifacts) does not equal need for IVTC, nor necessarily make IVTC appropriate for that video source. It might equal a need for de-interlacing, which is very different.

The elimination of interlacing artifacts is a side effect of IVTC when applied to film sources.

Are you trying to convert a MOVIE to 23.976 or just some video you recorded on a camcorder or off of live television? IVTC will work correctly for the former, but not the latter.

As for your last question, whether the output should be interlaced or non-interlaced - it will by nature HAVE to be non-interlace (progressive), since IVTC produces progressive video.

<font size=-1>[ This Message was edited by: kinneera on 2001-09-10 18:43:55 ]</font>

[homerpez]

Let ME try again...

Never mind. I think I figured it out...

I did AUTO IVTC, and got the feild order right, and it looks respectable...

It does seem to make a difference to IVTC rather than just 29.97 Progressive --> 23.976 progressive. I think it will go through and actually remove the "correct" frames (those with interlace artifacts...

I guess I can deal with that...

<font size=-1>[ This Message was edited by: homerpez on 2001-09-10 22:15:11 ]</font>

[kinneera]

The only point I was trying to make is that if your video was filmed at 29.97, every frame is interlaced and thus has the potential to exhibit interlacing artifacts when played back on a computer screen (or other progressive playback device). Thus, IVTC technically has no good criteria for eliminating frames. But if it worked for you and meets your criteria for quality, you can't argue with results!

[homerpez]

Actually, you CAN. They weren't CONSISTENT results...

I just did a clip the video I originally wanted to do IVTC (not my test clip)... and sure enough, I get BLOCKS. not just a few blocks here and there, but TONS AND TONS of them during motion scenes.

Apparently, I encode 29.97 --> 29.97 or give up trying... maybe a noise reduction filter? or two? or three?

I apologize, sometimes I REALLY know things of the forum and can help out people, other times I come off as my TRUE self, a total idiot.

[VidGuy]

For the Red frames -> They're the ones that will be used.

Typically, you'll see a pattern like 1001010010 or 0101001010 (where 1 is a red frame).

I've found that if you've got the correct field first, TMPGEnc's automatic IVTC normally does an excellent job on captured material from LaserDiscs and Videotape (if they were originally film, which is usually true for movies).

As someone mentioned, if the video was "shot" at 29.97, IVTC will be a waste of time.
I actually just create an interlaced SVCD - you are aware that you can just encode the video interlaced if you're making an SVCD - you don't have to deinterlace it. It will look crappy on your computer screen, but it will look great when played back onto your TV.

[kinneera]

What format is your source video in - MPEG, AVI, or other?

If it's MPEG, then it's possible that the blocks are a result of frame deletion, which I suppose in theory could corrupt the GOPs.

If it's AVI or other, then getting lots of block noise could point to a problem other than IVTC, such as insufficient bitrate or a low motion-search algorithm setting.

[digvid]

If you have software like Adobe AfterEffects, you can convert the 29.97fps interlaced video to 23.976fps progressive. This is about the only decent way I've found of making film-like video from interlaced NTSC footage.

- digvid

[homerpez]

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>
On 2001-09-11 05:25:34, kinneera wrote:
If it's AVI or other, then getting lots of block noise could point to a problem other than IVTC, such as insufficient bitrate or a low motion-search algorithm setting.
</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

Unfortunately, with this source video (Full screen, MPEG-2 5.35MB/sec 640x480), it apparently was just too tough a task... I HAD to, unfortunately, raise my bitrate for the encode. I had to go from 1170 AVG to 1670 AVG bitrate, before some of the blockiness went away... I guess this is allright, since I sould be able to fit 5 11-minute clips on one DISC this way... still not bad...

VidGuy: I did finally discover what the red ones were, and noticed there was a pattern to it... but even converting to 23.976 did not help the blockiness problem, and I had to still raise bitrate to 1670 AVG.

I guess I didn't want to resort to that because I was used to getting xSVCD's made from whole DVD-rips, at only 880-1000 AVG bitrate, using CCE and MPEG-2! And having it come out perfectly!

Apparently, there is a difference between full-screen NTSC 29.97 frames, and letterbox FILM 24 frames... lesson learned...