TMPGEnc Experts - what do you think of this?

[R_A_P_D_]

Hi, I am trying to improve my svcd profile (video only - I encode .mp2 separately).

GOAL: reduce blockiness as much as possible without raising the bitrate too high or slowing down encoding too much.

So far my best trade off PROFILE is below (see images):

Do any of you TMPGEnc Experts know how to improve it?
Perhaps I have made some embarrassing mistakes in it?

For example maybe I should make the
DC component precision 10 bits?
Or check "Detect scene change"?

I appreciate any help to reduce blockiness!
Thanks!

[mickboss]

CHECK:

- Detect Scene ...

UNCHECK:

- No motion search for still ...


Motion Search Precision:

- VERY HIGH

[R_A_P_D_]

Thanks for your help!

Now in my limited testing I have found on fast moving scenes "Detect Scene" hurts more than helps, oddly enough.

Lukesvideo seems to agree:
"2) Make sure you don't have the scene change detect option enabled. If
you leave this on, TMPGEnc may insert too many keyframes. This can
lead to the sudden change in quality from frame to frame that you
mentioned."

From:
http://pub30.ezboard.com/flukesvideoboardfrm5.showMessage?topicID=42.topic

Also I have read that Motion Search Precision: VERY HIGH is not better than HIGH, it just takes longer---do you agree?

Thanks again, not trying to be argumentative just trying to learn....

Any comments appreciated and thanks again...

[Triffid]

I'd use the 10 bit precision,

Up the bit rate a bit (I think you are actually BELOW standard...which can't be good) especially the average and minimum. If you don't mind using the space, go CBR. I find that the 2 pass thing tends to allocate less bitrate than you have said you will allow it to, even in the hard to compress scenes. Also I double the VBV setting, which I think helps.

I use manual setting of the scene detect stuff (under force picture type setting). I don't trust TMPGEnc to do it itself. I agree that more I frames means more quantization, but it can also help the picture quality if you put them in the right place. I set each scene change to "I, New GOP". At least that way the GOP pattern starts again each time, which may keep the ratio down compared to adding 'random' I frames in the existing structure. It seems to clean up the transitions very well.

One other thing to consider is whether you really need all that resolution. For my 'SVCDs' I use 1/2 frame size (like VCD). You don't loose that much res, or get that much lesssmooth flow in the video, but you sure get less blocks (much more bitrate per macroblock). It works fine on my pPioneer DVD player, but may not work for all.

Otherwise, I agree with the other good advise youve been given.

[$$$]

How about some letterboxing and increase the bandwith by not encoding what is not visible on your TV set? Instead of Full Screen in Video Arrange Method just Center it to 448 x XXX (XXX depending on your source).

$

[R_A_P_D_]

Well I updated the graphics to reflect the improvements I have made in the war against blockiness... The new settings are very much less blocky, though they work in conjunction with virtual dub filters to achieve this (blur and 2dcleaner).

Thanks to Mickboss for
UNCHECK:
- No motion search for still.

This helped.

Thanks to Triffid for vbv 224, which helped, and for telling me to up the bitrates that worked too (apparently 2pass vbr new lies about the average... on 2000 average the real average is 1850...)

10 bit precision seems not to help, however.

Thanks to $$$, though I do know how to letterbox, but won't that destroy
the aspect ratio? I capture at 480x480 from vhs so its not widescreen. What should I resize it to before adding the black bars to make it 480x480 again?

Any other comments appreciated...

[baker]

I ahve the answer to ure problem!!

u want low bitrate (120mins on 2 cds) with NO blocks and high quailty picture???(very close to dvd)

than use the cce encoder. there is a new guide just released.

Baker

[adam]

I have to address some of the suggestions that Triffid made. At svcd bitrates it will actually produce lower quality using dc 10 because it takes away bitrate from your picture.

His bitrates are well within the standard. As long as the video bitrate does not ever exceed 2.6mbits then it is compliant.

I would leave your min bitrate setting alone, at the most raise it to 500kbits. As far as the avg bitrate is concerned it all depends on how many disks you are willing to use.

There is absolutely no point in changing the VBV buffer size to anything other than the automatic settings. This option has no effect on quality at all it just sets the buffer size. Basically if you arent having serious playback problems (freezing picture, huge discolored blocks ) then your VBV buffer is fine.

R_A_P_D_ I don't know what the source is that you are encoding. If at all possible you should really do an inverse telecine or if your source is dvd then use force film. Encoding at 23.976fps (what you get if you IVTC) instead of the interlaced 29.97fps literally saves you %20 of your bitrate. Its basically like using a bitrate that is %20 higher.

[R_A_P_D_]

Dear Adam,

Thanks for your comments.
My testing agrees with you on the dc10 issue.

Concerning average bitrate, I want to fit 50-55 minutes on 1 80 minute cd-r.

Now concerning the VBV buffer, I slightly disagree. I have read the the buffer value is 112 for svcd, that is the standard. Yet it appears that TMPGenc.exe messes up because when i open a .m2v in Bitrate viewer (http://www.tecoltd.com/bitratev.htm) it says 56 vbv...even though the setting was 112 in TMPGenc.... when I double it in TMPGenc, then it says 112 in Bitrate viewer... maybe this is a bug in the encoder?
Though your probably right that 56 would work fine too...

What do you guys think of CQ (constant quality) mode? not vbr_Cq, just CQ... version 2.52 has both and both are vbr.... I find the bitrates increase considerably but the quality is quite good.

Last comment, I purchased both Ligos 3.5 and Tmpgenc 2.52 plus and so far I must say Ligos 3.5 is superior.... especially in handling high motion scences without blockiness... why does everyone say Ligos is inferior?

Thanks and any comments appreciated....

[adam]

50-55mins per 80 min cd is very reasonable. Simply use a bitrate calculator to determine the appropriate bitrate.

TMPGenc very well may have a bug in setting the vbv buffer, though I think its more likely that the bug lies in the bitrate viewer, but in either case my opinion on the matter remains the same. VBV is not a tweakable setting like many others in TMPGenc. It has no positive effects on quality it can only cause problems if set incorrectly, so if it ain't broke don't fix it.

All of the CQ modes in TMPGenc are strictly used to decrease encoding time. If this is your main concern then it is a viable encoding method for you but the fact remains that anything CQ or CQ_VBR can do, multipass vbr can do better because it analyzes the file first and then encodes, allowing it to allocate bitrate better. Furthermore, with CQ there is no way to predict filesize so often you get stuck encoding 2 or even 3 times to get it right, negating the whole purpose of using this faster encoding method. If you do choose to use CQ try not to use a quality setting much higher than %60. Beyond this point quality begins to top off while bitrate, and ultimately size, increase dramatically. At a level of 100 you will literally be encoding in cbr

You never mentioned anything about your source so my main suggestion to you is still to look into IVTC'ing or using force film. You can tweak your settings in TMPGenc all day but you will never find anything to increase your video quality as much as IVTC'ing does. That single process literally increases your video quality by %20.

[incognito]

I have been getting very good results by starting with the "NTSCfilm"
templates in tmpg. Basicly, as I understand it, it drops the frame rate
to ~24fps and ad some encoding info so that your DVD player knows to
display the video properly.

Anyway, the fewer fps means that mote bits go into each frame
and therefore these is much less blickiness.

[FranckV]

RAPD is right, I had to increase the VBV buffer size to 224 to get my SVCD to play correctly on my DVD player.

As far as what Baker is saying, I have been using TMPGenc for a while and recently decided to try CCE trial version 2.64. So far the different tests I ran have not got me convinced that CCE is better than TMPGEnc. The only advantage I found so far is that it is faster though it does not do
all the other things that TMPEG does (pulldown, resizes, detelecine, multiplexes,...).

[$$$]

RAPD, sorry, I'm not following things daily here. Since you capture @ 480x480 your target size should be 448x448 to save two macroblocks both horizontally and vertically. So, center & resize to 448x448 (aspect ratio should be preserved that way).

To speed up the encoding you could alternatively just blacken the edges of your video with a 16 pixels wide frame without resizing it (just 'Clip' accordingly and 'Center' in tmpgenc).

$

[R_A_P_D_]

Dear $$$,

Thanks for the great tip! Resizing to 448x448 and adding black borders to bring it back to 480x480 does improve blockiness in high motion scenes incredibly while keeping svcd compliance.

There is also an added bonus I cannot figure out yet. Namely, the television does not seem to display the left and right side black bars added in virtual dub... so it still looks full screen! Only a little bit on the bottom is blacked out and i have tried this on two tvs (one philips one quasar...) Theorectically, I should see black bars on the left and right screens of the tv but they are not there.... maybe the tv was cutting off some of the original anyway? It didn't look that way though in my tests...

Strange...

Anyway thanks again to $$$ and everyone!

[vhelp]

rapd,

from what I understand, tv custs some of the video (top/btm/lf/rt)
You will see some black on your left/right on some clips (pending
your source and your editing/D/L etc. ) but when played back on your
tv, (depending on how many pixels lf/rt) your tv will just cut them
off.

-vhelp