XviD Quantization type, H.263 vs MPEG; and GMC for hand recorded material

[carlaron]

Hello,

I need some input, knowledge transfer on a few codec settings; I have read a lot, and have an idea, what they do, I still cannot generalize, what benefits most likely for hand recorded (slightly grainy/noisy) AV material.

does it make sense to use the H.263 matrix in favor of the MPEG matrix as it smooths more, therefore giving better file compression; or does MPEG give more detail.

also, does GMC (Global Motion Compensation) really not do much good? I thought that shakey hand filmed footage could gain from that. what is your opinion on that?

[jagabo]

does it make sense to use the H.263 matrix in favor of the MPEG matrix as it smooths more, therefore giving better file compression; or does MPEG give more detail.

smooth = less detail, so they are the same thing. Use the h.263 matrix if you are going for small size and don't mind slightly less sharpness and detail. I found the difference to be maybe 5 percent in Target Quantizer mode.

does GMC (Global Motion Compensation) really not do much good? I thought that shakey hand filmed footage could gain from that. what is your opinion on that?

I've only run a few little experiments with GMC and saw very little difference in file size of Target Quantizer encodes (on the order of 1 percent, even with only panning or shaky shots). But Xvid's 3 warp point GMC is not supported by any set-top Divx/DVD or media players. So it's best avoided unless you plan only on watching your videos via a computer.

[carlaron]

thank you, that helps with the decision;

i do have a standalone; and have used it for some time, but now I am more going towards higher quality playback;
I already have a PC doing my DVB-S work; and in the future i think a PC will do everything.

but maybe I will leave GMC off anyways; if it is pretty useless.

[poisondeathray]

You also have the option to preprocess shaky handheld footage

e.g. deshaker in vdub, and various denoising filters

These will help to reduce your bitrate requirements for a certain "quality" level, and help with compression

[carlaron]

yap, i am actually just looking into that, but the filter that I decided on is crashing virtual dub... smoothiq

[TCmullet]

(I'm pasting these posts from the tail of a thread I shouldn't have posted this subject in.)

jagabo: Jan 6 2009: The Mpeg quantizer is a little sharper than h.263 quantizer. But it takes more bitrate and generates a little more DCT ringing artifcacts.

tcmullet: Jan 6 2009: (I knew I shouldn't have mentioned quantizers; I'm bordering on "hijacking".)

In all the discussions I've read, I haven't heard enough to help me decide which to use. I want to preserve picture quality; it's that simple. In my short time since serious Xvid use, I've used the default. I've used several mpeg encoders over the years since '01 when I got into mpeg capture, so I've known this thing called quantizers was floatin around in there. When folks say "mpeg q. is sharper than H.26x" (and H. is softer than mpeg), I say "softer or sharper than WHAT"? I do high bitrate mpeg2 captures of video. I want to preserve them as perfectly as possible, but use Xvid. For many things I use Q level=1 or 1.1. Some things (hi-motion sports) I've settled on Q=4 to 6 (4 for the game, 6 for postgame crap). It's probably good I used the default of H.26x. But for movies, I'm going to TRY mpeg. I suspect your comment about ringing was based on an assumption of running xvid by bitrate. In that case, I would expect degradation of some sort. You said it would be "ringing". Your and their comments (plus my own "encoding intuition") make me believe that with "Q level=a constant", the mpeg will give a sharper picture, NO increased ringing, but a bigger file, as detail takes more bits to encode. I'm going to take a signficant minority of a movie and code it both ways. Will compare both visual and filesize differences.

I don't want it necessarily softer OR sharper than the original... I simply want it to be as close to the original as possible (generally). I'll see how expensive it is after my tests. I've been pleased that often Q level=1 has still allowed a movie to fit on 1 DVD-R in many cases.

No answer absolutely needed unless you feel like it, but if I had a question now, it would be: Do you think I'll get ringing in the case where I keep Q level constant thru out movie and for both quantizers?

[TCmullet]

I have made a test, and am surprised at the results. I did a full 82 minute b/w movie both ways in Xvid. (FWIW, 720x344, "20 Million Miles to Earth".) I would think that if MPEG makes it sharper, it will have a higher bitrate for a given Q value.

With all other parameters constant (including Q=1.0),
H.263: 3.23GB
MPEG: 2.80GB !!!

I expected MPEG to be *larger* not smaller. Anyone know what's going on??

I extracted a couple minute sample from both clips to copy to my flash drive for playing on the Phillips DVP 5992. I am not sure I could tell a difference. I confess that my monitor is archaic and deficient, a 25" that is circa 1992 and rather worn. (And med-bitrate mpeg captures that looked okay on it in 2001 look terribly pixelated on a big TV in Circuit City.) Was the MPEG quantizer version supposed look a bit sharper? I kinda thought so, but it's so subtle that I wonder if it's a self-fulfilled prophecy.

I loaded 2 instances of Vdub, one with each sample. I blew a sample frame in both up to 400%. I kept switching back and forth, but still can't tell if they are really different. Maybe MPEG was sharper. Maybe MPEG ringing (the ghost-like halos around sharp objects) is worse. But I'm just not sure.

That wildly unexpected smaller file size for MPEG blows my mind. Any thoughts?

I'm running another pair of tests tonight. Only difference between this and the earlier pair is that I'm adding the grey-scale filter on both to elminate a slight tint from this b/w film. And by doing again (and getting similar results) I'll reduce the odds I made xvid config error(s) with the prior pair of runs.

Hey, I just had a thought. In Vdub, I can save a .bmp of a given frame from both. Then could I attach them to this forum somehow? (I've never considered doing something like that on a forum.) Then you folks who probably have much better resources (including eyes) could assess this for me.

[manono]

If you want it to be closer to the source DVD, don't use either one of the 2 XviD matrices, but use a decent custom quant matrix. Then you can drop down to quant 3-5 and still get better results than using those lousy matrices, mostly tuned for very high compression. Try the Fox Home Entertainment matrix, or one of the other high bitrate ones. AVIs made with the custom quant matrices won't play well in all standalones, but I think they will in your Philips machine (unless you get bitrate spikes it can't handle).

matrix%20collection.zip

[TCmullet]

(Manono, I'll respond to your kind help shortly.)

I see I CAN attach images! Nice that my test frame happened to be a key frame. This is the H263 one.

[TCmullet]

This is the MPEG one.

[TCmullet]

If you want it to be closer to the source DVD, don't use either one of the 2 XviD matrices, but use a decent custom quant matrix. Then you can drop down to quant 3-5 and still get better results than using those lousy matrices, mostly tuned for very high compression. Try the Fox Home Entertainment matrix, or one of the other high bitrate ones. AVIs made with the custom quant matrices won't play well in all standalones, but I think they will in your Philips machine (unless you get bitrate spikes it can't handle).

matrix%20collection.zip

I never thought I'd mess with quant. matrices, and now you're pushing me beyond my math abilities. Given that I don't know enough about DCT stuff to follow any discussion about them (nor do I need to, I think), a couple questions if I may...

1. How do I pick from over 60 matrices of which I know nothing and for which your kindly provided file has no documentation? You said "try Fox or one other hi-bitrate ones". Well, other than "best picture" or "high", most don't give a clue as to what they're best for. (Is there documentation somewhere?)

2. You say, "Unless you get bitrate spikes it can't handle". I do have a bitrate viewer, but haven't used it as so far movies encoded at Q=1 (H.263) haven't had playback problems. But I don't want to have ANY playback problems, even at whatever high rates the encoder uses when I'm at Q=1. I can't spend days "trying" 60 matrices, especially as I don't really know what to look for when comparing. (I really don't have much time for testing lots of matrices at all.) What am I do to?

3. Sad to hear the 2 standard ones are for "high compression" (great size reduction), whereas my goal is "use all the space I need in order to preserve everything". I'm tempering it with "get the movie down to 4.0GB". So even if you're saying I COULD get away with changing Q to 3-to-5, wouldn't I preserve info even better by staying at 1.0?

I realize these questions perhaps overlap, so can you address them as one? Which matrix should one use given my desire for best quality at (moderately) high size, and the need for perfect playback, too.

(BTW, I'm inferring from what I've read so far here and elsewhere that no matter what matrix I use, it still plays on anything that will play Xvid. I gather that the matrix is stored in the header by the encoder so that the decoder can play it. I wonder what this might imply for joining 2 AVIs that have different matrixes.)

[TCmullet]

I have found this bit of data on Doom9:

here is the lookup table
Test % -> Matrix
50 "Sharktooth's EQM v3ULR.cqm"
85 "Jawor's 1CD Quantization Matrix.cqm"
95 "Sharktooth's EQM v3LR.cqm"
180 "Sharktooth's EQM v3HR.cqm"
250 "Didees SixOfNine-HVS.cqm"
350 "Fox Home Entertainment.cqm"
450 "Sharktooth's EQM v3EHR.cqm"
over "Sharktooth's EQM v3UHR.cqm"

Should I infer from this that "Sharktooth's EQM v3UHR.cqm" is the best to use? (UHR meaing ultra high rate)

[manono]

1. I gave you a suggestion as to which one to try. Selur's are pretty good also. And Sharktooth's EQM v3UHR. The Soulhunters V3 is OK.
2. If using Q1 doesn't make your player stutter or freeze temporarily during the complex scenes, maybe you're OK. I don't have the player so I can't test.
3.

So even if you're saying I COULD get away with changing Q to 3-to-5, wouldn't I preserve info even better by staying at 1.0?

Not in my opinion, you won't, not if using the H.263 or MPEG matrices. Both basic matrices throw away so much information to begin with that even making Q1 encodes won't get it back. And all using Q1 does is bloat the size. It's a waste of time. If you're asking if using Fox Home Theater and Q1 will retain more information than using Fox Home Theater and Q3, the answer's 'yes'. But at the cost of a huge file size. I'm saying that you shouldn't even use Q1. The only reason it was ever put into XviD was because of newbies and other incompetents whining about XviD encodes not reaching the size requested.

Edit: I was looking for the Didee ones under Six-Of_Nine. I forgot they're named Didee... They're very good, but the Inter Matrix begins with a 10, which as far as I know is non-compliant. It'll definitely work on a computer. I don't know for sure about your standalone. You'll just have to test, I guess.

[TCmullet]

Thanks for your very helpful insights. I did find the Custom Quantization Matrix Editor, and it's interesting (though cryptic) to view the different shapes/slopes from the various matrices. But especially intriguing was your comments about the 2 std with low-Q. Yes, I haven't seen much difference between Q=1 to 3 or so, but I credited that to my poor monitor. I've been doing Q=1, then bumping it up (1.1 or 1.2 usually) if the movie ran over 4.0GB (and adding a bit of a "cleaner" filter).

I will definitely try Q=1 with "Sharktooth's EQM v3UHR.cqm" to see how much higher than the 2 std ones it goes. (I have a dedicated machine I can do all this on.) Will do these tests for at least this movie. Will pull out parts (high motion such as panning or otherwise lots of motion) to play on player via flash drive to test for some kind of player glitch. Then when I find a Q value to settle on, both for this movie and maybe as a starting point for others, then I'll switch to "fox" matrix to see how the size changes. (However, that "fox" matrix looks intuitively odd in the viewer; not as smooth a slope as with Sharktooth's.)

Hey, one last question... Perhaps I shouldn't have picked this movie for my test, as it's b/w. Can you assure me that all these quantization issues are irrelevant to whether the video has color or not? (I don't want to have to pick a matrix based on whether color or not.)

[jagabo]

matrix%20collection.zip

Thanks for that link!

[manono]

Hey, one last question... Perhaps I shouldn't have picked this movie for my test, as it's b/w. Can you assure me that all these quantization issues are irrelevant to whether the video has color or not?

I can't assure you positively that you needn't pick a matrix based on whether the movie is black-and-white or color. I can only tell you that because color movies don't compress quite as well, I may use a lower quality matrix when choosing one for the file size and quality I hope to achieve with a color movie (for DVD encodes). Other than that, I have never read of these matrices being tuned for color or b+w, and don't choose a matrix myself depending on whether the movie is color or b+w. In retail DVDs, the same matrices are used by the studios whether the movie is color or b+w. That Fox Home Theater matrix is commonly used in retail DVDs. I suppose the guy that gave it its name found it on a 20th Century Fox movie on DVD. Most of the matrices in that collection were created by XviD users several years ago, many with the help of that Custom Quantization Matrix Editor you discovered.

[TCmullet]

I've done some tests and here is my data.

H.263: 3.23GB Q=1.0
MPEG: 2.80GB Q=1.0
MPEG: 2.75GB Q=1.0 Max bitrate=17231 Avg=4443 kbps (For this one, I added the filter "greyscale". I did expect slightly smaller size, but wasn't sure if the eliminating all traces of color would matter. Guess it does, a little. Kept greyscale filter for all testing below.)

I'm still not clear how that if MPEG is supposed to be a bit sharper, it's taking less bits rather than more than H.263. However, this issue seems to be eclipsed by the option of using a "custom" matrix, which I've done. Here is the further data I have, using my b/w movie 20 Mil. Miles to Earth, 720x344, "Sharktooth's EQM v3UHR.cqm":

Q=1.0 5.85GB Max bitrate=28625 Avg=9831 kbps (Whoa!!)
Q=1.5 4.11GB Max bitrate=24415 Avg=6809 (I must stick to 4.00GB and smaller.)
Q=1.6 3.88GB Max bitrate=22613 Avg=6413
Q=2.0 2.86GB Max bitrate=16807 Avg=4647

I cut out a 2-min. segment to play on the Phillips from the Q=1.6 and 2.0; a segment with many peaks (fast motion). Neither caused my player to hiccup in any way that I could see. And I can't see any difference in quality. I'm inclined to go with 1.6 in this case, rather than 2.0. But if I stick w/2.0, the max rate will be comparable to the MPEG max rate. Perhaps this will make it compatible w/more players. Any thoughts, anyone?

One further question: My 2 test samples were played from a USB flash drive, which obviously has a faster access rate than any spinning disk. What with the non-sustained rate of some of these being over 20Mbps and the DVD standard sustained rate being 9.8Mbps, I'm wondering if the Q=1.6 to Q=1.0 files will have a problem due to the disk not being able to be spinned fast enough. I'm hoping things are transient enough that the player's buffering will be enough. I don't want to burn a series of DVD-Rs to find out.

I'm thinking of using 1.0 strictly for archival of non-movie critical footage (which I can split up among multiple disks). But for movies, it appears I must stay below 1.5 so I can keep each movie to 4.00GB.

Also, I don't see any reason to use any of the rest of the custom matrixes, as I'm not trying to squeeze everything to the tiniest possible size.

But I'm open to all suggestions.

[jagabo]

Yes, DVD players usually max out at ~10 kbps. If you have sustained peaks over that (for audio and video combined) you will have problems. Many players will not even handle 10 kbps when playing Xvid files. Divx Ultra certification only requires about 5000 kbps. Many older players can only handle about 3000 kbps. You're asking for trouble.

[TCmullet]

Thanks, jagabo, but my player DOES play those two short clips. However, it's from a USB drive, not a spinning disk. The question is whether a playback problem (of sustained or non-sustained) high rates can happen SOLELY due to disc-spin rate limitations (on this or any Xvid player), EVEN though the player plays the file fine when getting it from a flash drive.

manono, I'll greatly appreciate your input on this too, as well as your thoughts on all my collected data and what the implications are, from your perspective.

[jagabo]

I understood what you were saying and I responded specifically about playing off DVD discs.

[guns1inger]

For the DVD format, the maximum supported video bitrate is 9800 kbps, with a total combined bitrate of 10080 kbps (audio and video). As has been stated, many players cannot achieve this satisfactorily from burned media.

For Xvid encoded material, most players struggle with bitrates over 2000 kbps, although some will go higher.

Playback of anything outside the DVD format is not based on universal specifications, so each player supports file differently. Some players will not play MPG files, but will play VOB files. Some will play some formats from some media (eg. USB) and not from others. Some will manage higher bitrates than others. Some support different features to different levels (qpel, gmc). There are no standards, so all bets are off once you stray outside the lowest common denominator settings.

[jagabo]

My Philips DVP-5990 will play over 20,000 kbps from a flash drive. But it's limited to <10,000 kbps from a DVD.

[manono]

manono, I'll greatly appreciate your input on this too, as well as your thoughts on all my collected data and what the implications are, from your perspective.

I also thought jagabo answered the question. I haven't done any tests for awhile but from before, I had a player several years ago that choked on any XviDs that were over 4000 for about a second. It would stutter and freeze temporarily. The loader (the DVD-ROM part of it) gave out and I replaced it with a retail Lite-On and another test I did that maxed out at over 7000 played flawlessly. I have no idea of the max speed of the 2 although I suspect the Lite-On was faster (2x .vs 4x maybe?), which could serve to confirm what jagabo said - that the spin rate of the loader is important.

In your test, isn't the XviD encode near to the size of the source? If so, wouldn't it make more sense to keep the source unchanged? Or switch to archiving using x264 which compresses much better? Also, my recommendation when using those high-bitrate matrices wasn't to use Q1, but 3-5, depending on the matrix used. In the case of the Sharktooth's EQM v3UHR, I daresay you could use Q5 and not be able to tell the difference from the source. But I could easily be wrong, not having tested myself. Again, though, you shouldn't ever have to use Q1, especially not when using a high-bitrate custom matrix. You asked so I gave you my opinion.

[poisondeathray]

My Philips DVP-5990 will play over 20,000 kbps from a flash drive. But it's limited to <10,000 kbps from a DVD.

Interesting. I'm guessing it has to do with the rotation speed and read /transfer spead of the red laser when reading from disc?

[jagabo]

My Philips DVP-5990 will play over 20,000 kbps from a flash drive. But it's limited to <10,000 kbps from a DVD.

Interesting. I'm guessing it has to do with the rotation speed and read /transfer spead of the red laser when reading from disc?

I assume so. I tested with CBR files.

[TCmullet]

I understood what you were saying and I responded specifically about playing off DVD discs.

If you review your comment, you'll see that you specifically referred to DVD players, not discs. (As I said, my player does both discs AND flash drives.)

[jagabo]

You asked about disc, I answered about discs.

Why do you even ask questions like this when you can just test it out for yourself in a few minutes?

[TCmullet]

Yes, DVD players usually max out at ~10 kbps. If you have sustained peaks over that (for audio and video combined) you will have problems. Many players will not even handle 10 kbps when playing Xvid files. Divx Ultra certification only requires about 5000 kbps. Many older players can only handle about 3000 kbps. You're asking for trouble.

You refer to players and files, but not to discs. But lets not belabor that point; you *did* discuss discs later, as well as did the others.

[TCmullet]

I feel better now about trying some real discs. Will get back here after I try.

[TCmullet]

My Philips DVP-5990 will play over 20,000 kbps from a flash drive. But it's limited to <10,000 kbps from a DVD.

Interesting. I'm guessing it has to do with the rotation speed and read /transfer spead of the red laser when reading from disc?

This seems very likely to me.