DV avi encode at 23.976 for svcd or must it be 29.97 fps

[dsa44]

I have native Digital Video as AVI and have been successful in encoding to svcd and vcd with tmpegenc but at 29.97. Can I encode at 23.976 fps in hopes of increasing bitrate a little? And if so, do I just enable 3:2 pulldown, or do I have to do all that telecining stuff? (which I don't yet understand and may not be worth it) Or does a dv avi need to be at 29.97?
Thanks, I get more confused the more I read these posts!!

[adam]

you can encode at 23.976 but yes you will have to do an inverse telecine. Also you can only perform an ivtc if your source originated as film. I never capture my encodes so I don't really know anything about that but I believe if your capturing from television than you should probably be able to perform an ivtc in TMPGenc.

If you are confused by ivtc than just try it the easy way first, hopefully that will be good enough. Load your source and make all your normal settings. Set output to 23.976 and enable 3:2 pulldown while playback on the video tab. Under the advanced tab double click on the inverse telecine tab. set 24 for the fps. click on autoencoding and select "enable while encoding."

Now do a test encode. Let it run for a while and stop it. Burn a quick test disk on cdrw and play it. If it looks ok then you should be able to encode the whole movie like that but keep in mind that the telecining patter can change throughout the movie so there are no guarantees that the rest of the movie will look right. you wont know for sure until you encode the whole thing.

But you should definitely at least try this since it gains a substantial amount of quality to your encode.

[tacosalad]

3:2 pulldown and inverse-telecine aren't applicable to DV video, or any non-film source for that matter. But DV is especially sensitive to distracting visual effects if you change the frame rate.

But if you choose to record SVCD, you'll have to use 29.97 fps anyway. SVCD doesn't really allow 23.976 fps, unless you cheat.

[disturbed1]

Since you can't change to 23.976fps, you can lower your GOP.
The GOP controls the compression of frames. In TMPG the default setting if I=1,B=2,P=5.

Switching the Predicted frames to 3 will result in having an Intra frame encoded sooner. Intra frames are allocated more bits, so quality does go up, but so does file size.
This has the same quality result as encoding to 23.976.

[tacosalad]

I've got to argue with that logic a bit.

It's true that I frames use more bits than B & P frames, but VCD & SVCD have only a limited bit supply. Adding more I frames decreases quality by creating "crushed GOPs"--that is, a GOP with too few bits to satisfy the I-frame's requirement.

If you want to see this effect visually, encode a few seconds of video using the I-frames-only option to take it to the extreme.

[disturbed1]

<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 2001-07-25 09:35:48, tacosalad wrote:
I've got to argue with that logic a bit.
If you want to see this effect visually, encode a few seconds of video using the I-frames-only option to take it to the extreme.
</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>
Exactly, I-frames only is bad, not what I said.
Your GOP should be ~.5 secs I frame to I frame. This is where the increased quality conception comes from by encoding at 23.976. 29.97 should have a distance of 15, where as 23.976 should be 12. This is in no way a crushed GOP.
An encoder compresses by frames, not anything else.
GOP of I=1,P=5,B=2
IBBPBBPBBPBBPBBPBB
I which is an Intra frame, a key frame that is referenced, always contains the most info.
P which is a Predicted frame, compressed slightly, still referenced
B which is a Bi-directional frame, highly compressed, never referenced.
By shortening the GOP to 12
IBBPBBPBBPBB
I=1,P=3,B=2
This allows for more bits to be encoded into that GOP. The stream is still smooth due to the still included B, and P frames.

If your bitrate is set at 2000kbits/s, 1000kbits/s is reserved for each GOP. An encoder does not encode by seconds, it encodes by GOP. It assumes you have properly set your GOP according to FPS.

This is even evident in GOP analyses of DVD's, during high action sequences the GOP is shortened in this manner to allow a higher detailed reproduction.

This is the whole concept behind Force auto I frame insurtion. With a high quality encoder (Not TMPG), and this option selected, during high action scenes it will automattically shorten the GOP by insurting I frames (a new GOP) to allow more bits be encoded to the highly detailed scenes.

[hitechjunkie]

<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>29.97 should have a distance of 15, where as 23.976 should be 12. </BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

ya know, this is exactly the settings i've been using lately for my XVCDs. i had read somewhere that GOPs should be between 12-15 frames in length, so i decided to use a 12 frame GOP for my 23.97fps encodes and a 15 frame GOP for my 29.97fps encodes. just kinda made sense to me to do it that way (each GOP length is ~half the framerate).

[tacosalad]

We might be making some conflicting assumptions.

In the context of the original discussion, I assumed you were reducing GOP size from 15 to 12 without adjusting either the frame rate or the bit rate. Even re-reading the message, I'm not absolutely sure whether that's what you meant.

If the number of GOPs increases without changing either frame rate or bit rate, then obviously the total number of frames and the total bits (size) would remain constant. Hence, that would allocate the same number of bits over a larger number of GOPs (fewer bits per GOP). But this is evidently not based on the same assumptions you intended.

Although it's only partly related, TMPG is particularly vulnerable to the "crushed GOP" problem--as I think you hinted. If you enable its "detect scene changes" feature, you'll occasionally see a visible drop in quality immediately before a scene change because the encoder has truncated one GOP and allocate its bits into the next one.

But, as you also observed, that's not a flaw of MPEG itself.

[AP Enterprises Ltd]

Remember me? Hehehe...
Looks like a nice topic you got here.

Anyway... I have an .avi file, 480 x 204, 23.976fps. I want to make a SVCD with high quality, and I've done pretty much that so far but have not touched the GOP settings. The resulting MPEG2 file must be 29.97fps because my DVD player has a problem with 3:2 pulldown on playback, so I have no choice but to encode it into 29.97fps.

Is there anything in the GOP settings I could change to get higher quality video? I know you said that if it is 24fps, you should have a GOP size of 12 but since I am encoding it as 29.97fps, should I just leave it as is?

[disturbed1]

<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 assumed you were reducing GOP size from 15 to 12 without adjusting either the frame rate or the bit rate.
</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

This is true

<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>
If the number of GOPs increases without changing either frame rate or bit rate, then obviously the total number of frames and the total bits (size) would remain constant. Hence, that would allocate the same number of bits over a larger number of GOPs (fewer bits per GOP).
</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

Your partly right, the number of frames wouldn't change, but the size will.
An encoder does not encode by seconds, but by GOP.

Like stated above, if the bit rate is 2000kbits/s then 1000kbits is allocated to each GOP. 900 frames has 60 GOPs @ a distance of 15 60,000kbits would be allocated.

Or 75 GOPs for a distance of 12, w/75,000kbits in size.


Given the same number of frames, a video with a GOP of 15 will be smaller than one with 12, because there is more compression on the P and B frames.

GOP=15 has more P and B frames, and yields fewer overall I frames than a GOP of 12.

<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>
Although it's only partly related, TMPG is particularly vulnerable to the "crushed GOP" problem--as I think you hinted. If you enable its "detect scene changes" feature, you'll occasionally see a visible drop in quality immediately before a scene change because the encoder has truncated one GOP and allocate its bits into the next one.
</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>
Yep

The reverse can be done too. I've only tried it with SVCD though. I've encoded clips with huge GOP's and obtained great compression, with I=1,B=3,P=5 (24).

The standard for SVCD only says the GOP should have a length less than 2 seconds between I frames. VCD says ~.5 seconds.



<font size=-1>[ This Message was edited by: disturbed1 on 2001-07-25 14:58:24 ]</font>

[tacosalad]

I guess that's where our assumptions didn't mesh--the change in file size. For SVCD, a change in size would be feasible.

Don't forget, however, that the size of a VCD-style mpeg (at least a standard-compliant one) is determined solely by its play time. It was careless of me to extend this to SVCD, but force of habit I guess.

But of course we've completely diverged from the original topic!

[dsa44]

Diverged! ... I'll say, but what a great discussion! I have found that my dv avi does not play well at 23.976 with 3:2 pulldown. Too jumpy. Nor at 23.976 without 3:2 pulldown. Tacosalad may be right in that you can't really change the frame rate of digital video (camcorder dv avi source)

[tacosalad]

To be honest, I "borrowed" that observation from others who have discovered DV videos look fine as SVCD but not VCD. The conversion from 30 FPS interlaced (60 "pictures" per second) to 30 FPS non-interlaced (30 "pictures" per second) was adequate to make motion look annoyingly jerky--at least for many DV users.

If you're recordings are short (under 45 minutes), use SVCD and just crank up the settings!