CQ beats VBR bigtime...

[davidian]

Ok, I've not made any major tests, but I'm at it. I've done some smaller tests with video files, only with short clips, but now yet with a whole movie.

What I'm talking about is converting DVD to SVCD using CQ in tmpeg, Constant Quality. In my opinion, this gives the absolutely best quality.

Check out for example

http://tangentsoft.net/video/mpeg/enc-modes.html

Every time I've tried encoding a movie, people have been telling me to use 2-pass VBR or 3/4-pass (CCE) since it gives the best quality, and is predictable.

Well, I couldn't disagree more. Sure, CQ is highly unpredictable, but if you try it a few times (ofcourse, encoding one movie would take a very long time to get "perfect" results) the quality is much better than for example CCE SP with 4-pass VBR.

Now, I'm not an expert, I don't actually know ALL theory behind all this, a lot about mpeg whatsoever, but I have been experimenting a lot to see what gives me the BEST visual quality (and compression)and every time my conlusion is: Tmpeg encoder with CQ gives the best results.

However, I'm going to conduct several experiments with many diffrent movies, to see where it takes it. If anyone else's made similar tests, and can prove that VBR will give better results, please show me.

I made one test, a short one, with a scene from Terminator 2. This was a scene with lots of movement, and in my opinion, these are the scenes that affects the quality most. The "calm" scenes look quite the same with VBR and CQ according to me.

Anyway, first I thought there was going to be lots of macroblocks, but there were no macroblocks with neither CCE nor Tmpeg. (Not with LSX either).

There weren't actually any GREAT differences between the quality, but there were some small, and those matter a lot too. I'm talking sharpness.

Settings:

Tmpeg:

CQ - min 0 - max - 2000 - Q: 60
CQ - min 0 - max - 2300 - Q: 60

CCE:

4-pass VBR - min 1200 - avg 2300 - max 2600
4-pass VBR - min 1700 - avg 1900 - max 2600

The results were slightly smaller filesizes with CCE, but worse quality. Both tmpeg-files had better quality, and I believe the file with 2000 kbps as maximum bitrate was ~ the same filesize as the CCE files.

Anyway, I'm going to make some big tests with low-motion movies, high-motion movies and semi-motion movies, using CQ and VBR, to prove the differences.

Conclusions so far: To get good results with CQ in Tmpeg, you might have to encode the file many times to get the right settings, but since the quality/compression is considerably better than i.e 4-pass VBR in CCE (and ofcourse 2-pass in tmpeg!) it will still win, according to me. I hope that I will get the same results after doing these tests..

Any comments?

[maxqa]

You may be right but are you not missing the point ?! VBR is used to give the best compromise between quality and a predictable file size.

As you say "To get good results with CQ in Tmpeg, you might have to encode the file many times to get the right settings", well - who wants to spend time doing that !

[adam]

How can you compare CQ to VBR and use different settings for each encode? Your max min and avg settings need to be identical between encodes otherwise theres nothing to base your results on. Furthermore, your min settings in cce are WAY too high. With that high of a min that leaves the encoder with much less bitrate to work with, meaning your high bitrate spikes will be much lower, which translates into overall lower quality.

Try re-running those cce clips with a min set to 0 and you should notice a huge increase in quality and maybe a decrease in size as well.

[thxkid]

Hey I use nothing but CQ or CBR and I have perfect and I mean perfect backup copies of my DVD's, Laser Discs or Hi8mm Camcorder family stuff, my SVCD's have no macroblocks and I am viewing this back on a 36in TV and a 52 in Big Screen. I want no compromise, I am looking for the highest quality picture and sound I can get.
I have been burning VCD's for over 2 years now and what I have came up with (my own templates) gives me perfect VCD's.
Now I am switching over to SVCD's(higher quality).
I tried using VBR, takes longer without any or worse picture quality.
I only get 38min. per 80min. CD-R's on VCD's and SVCD's but the quality is worth it to me, I now get 48min. per CD-R's using the 99min. CD-R's.

[davidian]

Maxqa, it sure takes a lot longer time, but the result is that the quality is a lot better.
Adam, the reason why I chose that setting is because I had a similar discussion before. then, I used 600 as minimum bitrate, and got the same results - CQ was a lot better - so a person on this forum, Kdiddy told me to raise the min bitrate to at least 1200, or else the quality should be worse.. so I don't know which is true.

Since I can only choose min and max bitrate in Tmpeg with CQ, I can't specify an average rate. So if I use min 0 and max 2300, the closest I get in CCE should be min 0, avg 2300 and max 2600.. or what? I mean, to make a juste comparsion..?

I will try what you suggest, but I'm still quite sure the quality with tmpeg and CQ will be better.

[Kdiddy]

davidian: "so a person on this forum, Kdiddy told me to raise the min bitrate to at least 1200, or else the quality should be worse.. so I don't know which is true."

If you are going to drag my name into this, PLEASE quote me correctly, as I have this discussion before..I told you to raise it to 1200, but also, & more IMPORTANTLY to keep you range on either side of your min average the same....also keep in mind that the method I provided you works FOR ME, I did not gaunrantee you will like it as well, as quality is a SUBJECTIVE matter.

"CCE:
4-pass VBR - min 1200 - avg 2300 - max 2600
4-pass VBR - min 1700 - avg 1900 - max 2600"

This is NOT what I mean, I recommended that the max/min value be no closer than 500 bits, thus at least giving a minimal overall range of 1000 bits...using what you have written, what I told you would have resulted in this assuming 2300 is your desired bitrate...bare minimum would have resulted in 1800-2300-2800. However, I DO NOT apply this method to this high of a bitrate, because personally I find that CBR is acceptable for me on my DVD player at this rate.

"Sure, CQ is highly unpredictable, but if you try it a few times (ofcourse, encoding one movie would take a very long time to get "perfect" results)"

Well this is what will turn most people off fromy uor statement. To most people, they are shooting for the best possible quality in the shortest amount of encoding time. But trying a "few times" on a movie that takes 8-10 hours to encode with TMPG, is not what people are looking for...I agree with Adam, IF you are going to compare the 2, then you must use the same values...I personally find that using avisynth w/ minimum 3pass VBR CCE, or VFAPI wrapping a TMPG project file then encoding w/ minimum 3 pass vbr CCE, provides a beter quality product with less time than unpredictable CQ.

[adam]

Well there is another thread which goes into detail about how max and min bitrates should be set. You should read and decide for yourself what is best but I recommend always using 2.5 as your max and between 0 and 500 for your min, regardless of what method of encoding you are using and regardless of what your average is. But if you are going to compare CQ to VBR then you must use the same max and min settings for both otherwise its an unfair comparison.

The only way to do a fair comparison between CQ and VBR would be to only use TMPGenc since cce doesnt have a CQ mode. But to compare CQ and cce's VBR you would again need the same max and min settings. It would be best to frameserve to both programs and do resizing and such like that, otherwise you dont know if the quality differences are because of the encoding method or the encoder itself. Since you can't set an avg in CQ mode the TRUE way to test it would be to first encode in cce and then to keep adjusting the CQ setting until your file matched the size of cce's. Of course this is not how CQ is meant to be used.

CQ guarantees quality because even with a low quality setting it is very generous with the bitrate. Your quality will be good but the size can often be very large. If this does not bother you then CQ is fine since it is faster to encode than vbr. But VBR encoding can match CQ's quality in a much smaller filesize, at the expense of encoding time. CQ's only real advantage over VBR is encoding time, if you are having to encode multiple times to get the size right then your probably better off just using VBR in the first place.

[davidian]

Kdiddy, OK sorry if I misunderstood. Anyway, I wasn't talking about what's most suitable for most people here, I was talking quality. I don't mean to make you angry or anything.

I made this test again with tmpeg vs CCE, using min 0, avg 2300, max 2600 like adam said. The quality indeed improved, and it was harder to see any significant difference between Tmpeg (CQ) and CCE. But, the quality with CQ was slightly better, and, the file size was smaller. The file size was almost exactly the same for the 2000 kbps file, but the 2300 kbps file was bigger (~6%). Ofcourse, I do realize that this test was not sufficient to clear this issue, since I did it on one 11 seconds long clip, with high motion. Like I said, I'm going to make some more tests with whole movies.

PS. Encoding a 100 minutes long movie with tmpeg (CQ) takes 29 hours for me But sure, I'm willing to do that several times, to get the best quality achievable.

[Kdiddy]

you didnt make me angry...just dont like being misquoted...29 hours!?!?!..ouch...Id like to do something though here, I want to encode the same scene as you, and send it you and have you compare it??...we are talking mpeg2 right?

[teamhawaii]

in light of VBR.. for someone encoding an SVCD.. does min/avg/mac make that big of a diff?

I usually go with 0/2100/2500... what I am curious is how the min rate affects everything.. will upping the min rate give better quality, greater filesize, hmmm?

[Kdiddy]

teamhawaii: that subject has already been tossed through the mud.

http://www.vcdhelp.com/phpBB/viewtopic.php?topic=63503&forum=11

<font size=-1>[ This Message was edited by: Kdiddy on 2001-10-18 17:38:22 ]</font>

[adam]

teamhawaii: To summarize, yes min, avg, and max settings make all the difference in quality and size. Raising the min will not raise quality it will in fact lower the quality. The only time it would help would be if the encoder had flaws and allocated too little bitrate to low motion/detail scenes. If your using cce or TMPGenc I dont think this ever really happens, at least in my experience. Raising the min setting should not increase size significantly. In a perfect encoder it wouldnt raise it at all, but of course there's no such thing.

Your avg setting directly affects quality and size. Keep raising it til you've maxed out the # of cds you want to use.

The higher the max bitrate the better, but beyond a certain point it doesnt do anything unless you raise your avg. Also theres only so high you can go before dvd players start having difficulty playing it.

Teamhawaii your settings are good, I wouldnt change a thing.

[teamhawaii]

thanks for the tip.. i kinda figured as much..

[mole]

Davidian said (The "calm" scenes look quite the same with VBR and CQ according to me.)
Ok - but with VBR, "calm" scenes are encoded at a lower bitrate, saving filesize, so that bitrate can be used for higher motion scenes. Why waste the bits on CBR? Or am I missing something? I understand that sometimes the picture can be degraded with VBR as it tries to boost the bitrate for higher motion, but (at least from my experiences) the more passes, the better this compensation adjusts. I usually use 300 as a min, and 2500 as a max, and adjust the average depending on how many discs i want to put it on. For movies with little action, i have fit up to 105 mins on 1 disc, and been VERY satisfied with the results, if there is a bit more action in a movie with the same length, i would adjust the average bitrate and split it on 2 CDs. Its all a matter of opinion i suppose, just throwing in my 2 cents.

[shochan]

It all depends on what you want, He likes cq I like 2passvbr. He likes noodles I don't. If it works for you and makes you happy then go for it. Here is a real test

Encode with your cq medthod you prefer. write down the file size.

lets say it takes 2 80min cds and fills it max with the cq with no room to spare.

I bet that if you do it 2passvbr and set it up with the same exact size as an output it would look better than cq using the same max and min. Its just common sense think about it.

Shochan

[davidian]

Adam, I'm not only looking for the best quality, but also the best compression. What I hope that my tests will show, is that CQ can give better quality AND smaller filesize, otherwise it would be completely pointless.

You are saying here, that VBR matches the quality with smaller filesizes, and if that is what is true, I would use always VBR. I'm not sure what is really true, perhaps you've made a lot more tests than I have, but I will continue testing this, because my opinion is that CQ should have better quality and smaller filesize.

For anyone that didn't know, I'm talking mpeg-2 SVCD here, only. Nohing else. And I'm talking about fitting one movie on 2 cd's, that's what's important for me.

[davidian]

Yeah I kind of realized I missed to mention the point of all this. Like I said, to fit a movie on 2 cd's (unless it's too long). So I claim CQ would do the trick best, and perhaps after trying a few times with each movie, find out the way to make the quality as good as possible, raising the bitrate until it fills 2 cd's out to the max.

Are you positive that CCE would do the job better? (3/4-pass). I'm talking about 2 80 min cd's, so the final file size would be ~ 1400 mb.

Oh yeah, that reminds me about a question that perhaps adam can answer, since you seem to be very experienced.

I encoded a movie with tmpeg, which resulted in a file that was close to 1400 mb, I believe it was 1380, and then I split it into 2 parts, and since I didn't use CBR, ofcourse, it was kind of hard to find out at which point half of the file was. First I cut the file with tmpeg at about 50% of the time. Obviously, there were more bits located in the later part of the movie, since the first half of the film was much smaller than 700 mb, perhaps 600 or 550 mb (whatever).

So, I told tmpeg to cut earlier, and the results were better, but I still didn't have a file that was close to 1380/2 mb. I tried cutting it earlier, and at last, I got a file that was 697 mb, and I thought that was great!

My logic told me, that if I cut the file from the same point, the second file should be 1380-697 mb, but hell no. The file turned out to be bigger, and my two halves, although the file was cut in the spot where the first half was ½ of the total file size, together were bigger than 1380 mb. I think that the problem was something like that, and it was really bugging me.

So, when I cut a file, does it takes a few extra mbytes or what? Hmm I don't know if anyone understands what I mean

[stanwebber]

if you ask me by making filesize i.e. being able to fit everything on 2 cds your primary concern, you've not relegated quality to the backseat, but rather are dragging it along with a rope tied to the bumper

[davidian]

Well if the movie isn't more that 2 hours, say it's 90-110 minutes or something, you sure can fit it on 2 cd's with excellent quality. If you put it on 3 cd's you might just as well use CBR (...)

[adam]

Davidian, yes I am quite positive that multipass vbr can achieve the same quality as CQ in a smaller filesize. The reason that I am so positive is because they are essentially doing the exact same thing, they are both utilizing vbr encoding except that vbr does it in multiple passes, which allows it to make better decisions as to where to allocate the bitrate.

Both cce's VBR and TMPGenc's CQ modes use a quality setting. This is the amount of degragation that the encoder will allow when allocating bitrate. It works on both ends. It will raise the bitrate of a high quality scene til it surpasses the acceptable level of degragation, and on low motion/detail scenes it will drop it as close to the level of degragation as it can to save bits.

The difference between the two encoding modes is that CQ only uses one pass, so as to be expected, it is very generous with the bitrate. If it weren't then it would defeat its purpose, since CQ is specifically meant to ensure quality at the expense of compression. CQ is essentially an automated version of cce's 1 pass VBR. CCE lets you specify an AVG bitrate, this is specifically done to to make an attempt at controlling the size, even at the expense of quality is you so choose. CQ has no AVG bitrate setting, it ignores compression and simply makes decisions based on a quality setting. So multipass vbr does exactly what CQ does except that it analyzes the file first and makes better decisions.

Multipass VBR can definitely achieve the same level of quality as CQ in a smaller filesize, but it might not always be easy. CQ encoding is easy, with a quality level of 60 your movie will almost always look great. In order to get multipass vbr to look as good you need to choose a logical average for your particular movie.

Now that is not to say that your CQ encoded movies will look better than cce's multipass vbr encoded files, to YOU. But if that is the case then it is due to the encoder itself, not its encoding mode since multipass vbr is a superior method compared to CQ, in terms of both quality and compression. Some people prefer the quality of cce and some prefer TMPGenc, the only thing for certain is that there is no clear winner, they are both very good.

As far as your question, sorry I got to go to work I'll try to answer it later.

<font size=-1>[ This Message was edited by: adam on 2001-10-19 12:42:25 ]</font>

[davidian]

Ok then, thanks for the answer adam. In other words, theoretically, 3/4-pass VBR SHOULD always look better than CQ (or the same but with smaller filesize).

I will just try this and make some tests, and see what suits me best..

Just one more thing. My experience from Tmpegs 2-pass VBR is horrible, I always get a lot of macroblocks, and compared to TMpegs CQ I always get a lot worse results. Shouldn't it be the other way around?

[adam]

Well I bet your CQ encode is significantly larger than the 2-pass encode though right? If this is the case then your average wasnt high enough when you did the 2-pass encode. TMPGenc's 2-pass encoding is very good in my opinion. You shouldnt have any problem matching CQ's relative quality level and in a smaller size, certainly at the same size. But if you encoded with an AVG of 2300, which you stated earlier then you shouldnt be seeing any macroblocks. There might be something else to your settings causing this or it might just be the nature of the source clip itself.

In regards to your problem...There is a great deal of overhead added to svcd streams when they are multiplexed together. From my experience TMPGenc doesnt add this overhead as it encodes, so the resulting file doesnt have it, which doesnt necessarily make it non-compliant but it does make it less compatible with dvd players. But if you demux then remux, or cut your file in TMPGenc then it will add this overhead, which is why your filesizes got larger.

Now to make it easier on yourself don't use TMPGenc to cut. Like you said its trial and error to get the filesizes right and also, from my experience TMPGenc either doesnt add SVCD scan offsets or it does so incorrectly, which causes problems on many dvd players. Use bbmpeg instead. It properly adds the offsets and creates a completely compliant svcd mpg and it also gives you the option of cutting by MB, which insures that you completely use up all the space on your cdrs.

[Ciovo]

Excuse me, how can you say CQ is better than 2 pass VBR ?!?
Do you see the difference with your eyes? How can you understand little differences ?!? What kind of apps do you use to do that ?!?

Thank you!

[davidian]

Ciovo, I'm not done with my tests yet, and I'm going to make a lot more. But I use my own eyes, and then I state: which has better quality. That's what I do.

Adam, some of the tests I made were some time ago, and I am going to make more of them, and what's most important right now, is that if I compare two files encoded with different modes, I will compare file sizes.

I suppose two files with the same lenght and same file size should have same avg bitrate, right?
I'll try it out some more..

[adam]

Ciovo the only way to do a non-subjective test of quality is to use a bitrate viewer.

You compare the Q value, basically the level of quality throughout, to your bitrate curve.

You want your bitrate curve to be as close to the Q value as possible. If your under it then your picture is degrading due to lack of bits, if your over it then your wasting bitrate.

I think you'll find that with a high enough quality setting CQ will for the most part put its bitrate curve well over the Q-value, resulting in a good quality picture. Multipass vbr however will still have its bitrate curve over the Q value but it will follow the Q-value more closely, saving excess bits and resulting in a smaller filesize yet with the same level of subjective quality.

Of course this is just a measure of the allocation of bitrate. Assuming everything else is constant then yes you can use this as a measure of outright quality. But when comparing files encoded from different encoders it all boils down to personal judgement.

[davidian]

Adam, I got some strange test results. I used the same settings in tmpeg and CCE, 4-pass vbr in CCE and 2-pass in tmpeg.. the file with tmpeg was a lot better. Much sharper. I didn't expect that?

[teamhawaii]

adam..

what if your Q curve is way higher than your bitrate curve.. for some reason, whatever I try, I can get them to close...

as stated my mpeg2 settings are:

2pass, 0/2100/2500...(min/avg/max)

[randallc]

In absolute terms, CQ gives higher quality output than multi-pass VBR.

When you're encoding a movie to fit on cdr's you loose the highlight ability of CQ because you're limited to the max bitrate for SVCD's of around 2600.

The beauty of CQ is that it is a constant quantization regarless of bitrate. This means that no matter how fast the scenes are or how complex the picture is, its always going to look great because it just throws more bits at it to get the job done. But when you impose a max bitrate of 2600, now you just neutered CQ so that it can't get the job done as it is supposed to. Say a particutlar segment of the mpeg wants 5000bitrate, but you limited it to 2600 - its not going to look as good as CQ could otherwise make it look.

For this reason, and the unability to know for sure what the file size will be with CQ, I suggest that 2Pass VBR for TMPGEnc is the only way to go if you are trying to keep the file size small.

Randy

[adam]

<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 2001-11-07 00:44:59, randallc wrote:
In absolute terms, CQ gives higher quality output than multi-pass VBR.

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

This really isnt true. Both modes are vbr and they are both working off the exact same algorithms. They are essentially the same if you use comparable settings. The difference is that with multiple passes the encoder can make more logical decisions as to where to allocate the bitrate in order to save space.

Either mode can achieve the maximum level of quality for a svcd. CQ %100 and vbr (min 2.6, avg 2.6, max 2.6)are both essentially cbr 2.6, which is as good a quality as you can get as far as bitrate is concerned.

Literally, multipass vbr can achieve the exact same level of quality as CQ but in a smaller filesize. If size is of no concern and encoding time is, then CQ is the logical choice. If you want the best balance between quality and compression than multipass vbr is the best choice, but either can achieve the same level of quality.

[adam]

teamhawaii: Sorry it took so long to reply, I forgot about this thread. Don't be overly concerned with getting your bitrate curve over the Q value. This type of anlysis is good for comparing one encoding method to another, not in judging outright quality of a single encode.

Your Q value will vary drastically depending on your source. Sometimes it may be impossible to get the bitrate level at or above the Q Value without using an extremely high bitrate, which is impossible if you want your svcd to be compliant.

What you should really be paying attention to is that the peaks and valleys of your bitrate and your Q value are proportional. That is the measure of a good encoder and ultimately a good encode.