Could VBR ever be better than CBR??

[SatStorm]

Hi Matt,
well, some quality freaks outhere, are ready to do anything just to push the encoder to it's limits! This method is an alternative, so why not mention it? It doesn't means that I support or suggest it, but I just mention it.

My opinion is, that using 2pass VBR with the same average/maximum is a waste of time. It is like CQ @100% in a way. But I know at least one person who do it because he believes that it boost a bit the quality! (no, I don't talking for me).

Finally I am not a moderator just an user of this forum, so I can't check IPs etc. But I use 4 different dial ups myself, 2 of them through satellite (2way, no europe on line), so the point is, it is very easy to have 2 - 3 nicknames and nobody can find you. There are also some sites out there, offering somekind of online browsers for totally anonymous use...
Anyway, this is another subject

[LeeBear]

D_Head: You just keep repeating yourself over and over again without putting up any argument to support what you think is right.

I ask you again have you gone and done the TMPG test I described? You never have explained why doing 2 encodes both using a CBR of 2520 but with different motion precision settings in TMPG gives you different results. Why is that? Why does the one with a higher motion search presision look better? Why even have a "motion precision" option?

To answer the original question if VBR can produce better quality then a CBR. Yes it can even at a lower average bitrate and the same max bitrate. I will explain how. Take a simple example of 3 frames one complex, one simple (for instances a completely black screen like during a cut scene), and one average. Now if you did a CBR with a rate of (2520 kbps) each frame of that sequence would take up approximately 84kb even though the simple frame doesn't need all 84kb since it's pretty simple to encode. So the total bits for those 2 frames would be about 252kb. Now if you did a VBR encode doing the same 3 frames, the encoder may need to use the max bits for the average and complex scene, 84kb, but it knows the simple black screen doesn't need 84kb to produce the same quality (it's just a black screen) it knows it can be encoded perfectly using only 12kb. So now your total bits used to encode the same three frames is only 180kb (60kb per frame average) to get "the same" quality. We have just shown how VBR can give "the same" quality as CBR at a lower average bitrate.

Now you add in the superior motion precision search that the extra passes gives and your "same quality" VBR at a lower average bitrate becomes a "better quality" at a lower average setting.

I think that's what everyone who posted in this thread is trying to say except for D_Head and his doppleganger.

-LeeBear

[liquid217]

ahhh, i love a good fight on the forums, but just for arguments sake, i thought i would contribute my 2 cents. First off, instead of hammering each side with ramblings, i think time (and braincells) would be better if you simply broke it down what you are trying to say.

D_Head: You appear to be saying that when using multipass vbr, that the extra passes are only collecting data (analyzing) the file for bitrate allocation.

Satstorm and Adam: They appear to be saying that the encoder is actually doing more that just allocating the bitrate, that it is actually refining the encode with every pass.

I would very much be interested if those involved in this thread could actually dig up information conserning multipass vbr, and acttually give details.

[SatStorm]

What more details to show?
I wrote my configuration during the tests, so you can test yourself! What more to do?
Sorry, but I can't understand the "more". If you mean upload samples and photos, then no, I am against this method. The only way is to test and judge yourself! Believe only yourself. I easily can say that I am the president of USA and also publish a photo of me living in the white house. What that means, that this is proof?

Also, there are many users here, which don't seperate the technical reality from the book specifics. The books are ok, so the specifics, but the tools we use are far away from being close to perfect, so the results the produce when we use them. When I report something, I determine with which tools I have those results. If I use other tools, I might have other ones. And all those results might differ from what the book says.

Leebear- I dont do your test, as I know the results it will produce.
Like ADAM I think you are missing some the fundamentals of Mpeg compression.
You have a good surface knowledge but thats about it.
Visit the berkley pages on the Net....they are very good.


Why do I keep repeating myself as I believe I am right.

Your Words.......

why doing 2 encodes both using a CBR of 2520 but with different motion precision settings in TMPG gives you different results. Why is that? Why does the one with a higher motion search presision look better?

Put simply,its searching with more prescision, looking for more subttle changes.......

Why even have a "motion precision" option?

read above

To answer the original question if VBR can produce better quality then a CBR. Yes it can even at a lower average bitrate and the same max bitrate.

no this was the original question Leebear, please dont move the goal posts.

If my VBR encoding has a maximum bitrate setting of 2520, and my CBR setting is also 2520, I can't see how VBR can be better, since with CBR you're giving it the maximum bitrate all the time.

and........

I think that's what everyone who posted in this thread is trying to say except for D_Head and his doppleganger.

when we get back to the original question even Satstorm see´s things the same way as me..

these are his words not mine

My opinion is, that using 2pass VBR with the same average/maximum is a waste of time. It is like CQ @100% in a way. But I know at least one person who do it because he believes that it boost a bit the quality! (no, I don't talking for me).

Cheers
Matt

[joes]

Well, the whole point is, VBR save the space for more stuff and try to do that without lossing to much quality, it will never be better than CBR equal to the MAX rate, if U have enough space CBR will be the simplest way to max the quality.

How to set VBR parameters to ensure best balanced quality - file size is sure a big issue and think we have a lot experinces to share.

But to allege VBR can be better than CBR2520 seems to me like saying that by to look after the best (or the right one for the sake) bargain for the same article U will get a better one.

I know that many of you disagree with me, but I can't understand why.

I also wish we can avoid words like shit in this group, sorry D_HEAD.

Hey Joes are you not just another alias of D_head........ 8)

[joes]

If CBR in multipass mode with one option do better than another, is it because the coder is able to alocate bitrate to the part of picture with most motion, when this is true (I don't know) we should selebrate and pay our tribute to those who wrote the prog. Actually it is doing the same within single frame as VBR do with the whole file.

But VBR will still be no better.

[joes]

After the last post of LeeBear the discussion became more intressting to me because it raised the question as following:

CBR2520 = 84 kb per frame.
VBR2520max = 84 kb per frame at max.

That's what I think it is. VBR is just finding out now much reduction each frame can take befor the quality deterioate too much.

Can some body tell me for sure (not what U think) if it is alowed to allocate data bits from one frame to another? If this is the case, how can one ensure the max rate at any time throughout the entire file? I get really interessted.

I want to limit VBR max to 2520 whitin the SVCD standard.

Liquid 217

D_Head: You appear to be saying that when using multipass vbr, that the extra passes are only collecting data (analyzing) the file for bitrate allocation.

What I am saying is that there is no point usin VBR when The Max and average Settings are the same, this goes against all logic. When this is the case CBR will give you the best result in a much shorter time.

For my SVCD`s I always use a VBR whether CQ or (x-pass, i use this the most). The reason I mostly use x-pass is it is easyier to setup (even though it takes double the time to encode)
By setup I mean, I dont need to make any test encodes to calculate the final size.

Yes Liquid 217 you are right when you say multipass vbr uses the extra passes to collect data(analyse)the video(or audio) to get a more acurate bitrate allocation, which is why VBR-xpass only makes sense when you set MIN,AVerage & Max realistic values. i.e Not average 2520 & Max 2520 as how can x-pass VBR optimise bit alocation for this, you wont even get a smaller file size?

cheers

[adam]

After the last post of LeeBear the discussion became more intressting to me because it raised the question as following:

CBR2520 = 84 kb per frame.
VBR2520max = 84 kb per frame at max.

That's what I think it is. VBR is just finding out now much reduction each frame can take befor the quality deterioate too much.

No. Bitrate is ALLOCATED per second but it is DISTRIBUTED per frame within that second. Even when encoding in CBR you are not getting an even amount of bits per frame, this is the whole point of motion searching.

People are talking about basics of mpeg encoding in this thread but no one is actually looking at this from those simple perspectives.

How does an encoder encode? You specify a series of I, B, and P frames. I frames have their function but aren't really relavent here. P frames are predicted, in determining how much bitrate to give this frame the encoder looks at previous frames. B pictures look both to past frames and to future frames. This is what motion searching is, the encoder analyzes other points in the source to determine how to distribute the bitrate, YES even in CBR. CBR is essentially a collection of little VBR encodes every second. The avg bitrate every second stays the same but within that second it fluctuates from one frame to the next according to the motion priority of that frame. The higher motion search precision you set in the encoder (keep in mind not all encoders even have this option) the farther backwards and forwards the encoder looks to get this motion detail.

Now when you encode in VBR what exactly is the encoder doing? Well it essentially first runs a CBR encode to gather information about the source. One of the main things it looks at is motion. It stores this information, and if you do more than 2 passes it updates this information with each pass. It then uses this motion information (other info as well) and uses it to ALLOCATE the bitrate to each second and then to DISTRIBUTE the bitrate to each frame within each second.

Do a 2-pass encode in TMPGEnc and read the log, what does it say?

frame= 36 ( 18.0) Bitrate: 2469 kbps Motion: 7.73 MotionScale: 0.00 CQ: 75.07
frame= 54 ( 18.0) Bitrate: 2514 kbps Motion: 7.10 MotionScale: 0.00 CQ: 75.07
frame= 72 ( 18.0) Bitrate: 2428 kbps Motion: 6.31 MotionScale: 0.00 CQ: 75.07....

Sorry, it doesnt fit but you get the idea. The encoder has calculated the degree of motion according to its motion search algorithms. There would be a value for motionscale as well but the sample I used was very short but you can see this for yourself in your own encodes.

With multipass vbr the encoder gets two cracks at determining the level of motion in each second of your film. It is the motion searching which determines the distribution of bits even when the avg bitrate per sec is the same (CBR.) CBR is really no different than VBR except that it has stricter limitations on it. With VBR your bitrate throughout the entire encode has to average out to a specified amount, but each second's worth can have any value. With CBR your bitrate has to average out to a specified amount every second.

This is as technical and authoritative an explanation I can give. All of the evidence to support this is just basic facts of mpeg encoding and can be found anywhere in any book on mpeg. The best way to prove this for yourself is to just try it out for yourself. I find it hard to believe that there are people in this thread who are arguing so adamantly against this logic without even trying it themselves.

Joes-

After the last post of LeeBear the discussion became more intressting to me because it raised the question as following:

CBR2520 = 84 kb per frame.
VBR2520max = 84 kb per frame at max

This is a rather bad example(I did not use The S--- word)

as it doesent take into acount the Mpeg basic structure......ie we dont just use I-frames. You cant alocate the same amount of bits per frame.

e.g. ..berkley


Level I P B Average
30 Hz SIF @ 1.15 Mbit/sec 150,000 50,000 20,000 38,000
30 Hz CCIR 601@ 4 Mbit/sec 400,000 200,000 80,000 130,000




Easier would be to say...

Have a look at this page it sompares CBR & VBR in a simple graphical way
link

http://www.tecoltd.com/enctest/concepts.htm

Hopefully you will see why I say for VBR to its important to pick meaningful values for Min-av-Max.


When you are setting your Min Max & Average parameters, take in to consideration this simple Bits per pixel Example for SVCD, CVD etc

PAL Res F/R Pixels/s Relative to DVD

DVD 720x576 x25 =10368000 100.00%
SVCD 480x576 x25 =6912000 66.67%
CVCD 352x576 x25 =5068800 48.89%

If for example DVD uses 5.0 mbit/s then a CVCD only needs 2.5 mbit/s to match DVD in pixels/sec.


Cheers

[/url]

[SatStorm]

Arghhhh.....
That tecoltd again!

I am out of here, it's pointless reading this one...

[joes]

Thanks ADAM, for explaining in detail, but I'm still woundering if what U said is true, rules are rules, doesn't the 2520 max set for SVCD means that the standard never accept one single time point at which the Data exced 2520 whitin a second? Will 29 frames at 84kb followed by one frame at 85 kb be leagal?

If CBR is in fact just smoother VBR which was your point, will many frames in both CBR and VBR inevitable be more than 84kb, how to prevent the data amont from to exced 2520kb throughout the entire file, in 1/30 seconds resolution?

Joes you need to look at the basic structure of MPEG

For example



Level-----------------------------I----------P --------B-------Average (BITS)

30 Hz SIF @ 1.15 Mbit/sec 150,000 50,000 20,000 38,000
30 Hz CCIR 601@ 4 Mbit/sec 400,000 200,000 80,000 130,000

[joes]

Thanks D_HEAD too, I looked at that page, as I understood, if you super impose the CBR fig onto the VBR one, the horisontal CBR green line will limit the top of the VBR green line, when the VBRmax is set at the same level as CBR rate.
Is that right?

[joes]

But now UR talking about the uncopressed stream D_HEAD?

I'm sticking to one single issue, can VBR_max_x ever be better than CBR_x in picture quality?

I woudn't invol PAL/NTSC or DVD/CVD......, just not to be to complicated.

Yes exactly Joes,

Now do you see why making Max & Av the same for VBR is pointless.

Cheers

Matt

[joes]

Hi D_HEAD, I've never questioed that, will not set avrg=max anyway.

I'm leaving. Thanks to all for disc..


Joe

[adam]

Thanks ADAM, for explaining in detail, but I'm still woundering if what U said is true, rules are rules, doesn't the 2520 max set for SVCD means that the standard never accept one single time point at which the Data exced 2520 whitin a second? Will 29 frames at 84kb followed by one frame at 85 kb be leagal?

If CBR is in fact just smoother VBR which was your point, will many frames in both CBR and VBR inevitable be more than 84kb, how to prevent the data amont from to exced 2520kb throughout the entire file, in 1/30 seconds resolution?

This is basic math. If the total bitrate used in all frames per sec cannot exceed 2520kbits (this is the specified limit in this example, the actual SVCD limit is 2600kbits) than how can any one frame exceed 2520kbits? The rest of your frames would have to have negative bitrate which obviously isn't possible.

Again, bitrate is allocated per sec. So if you use CBR 2520kbits than that will be divided up over ~30 frames every sec (NTSC) but not all of those frames will have an equal amount of bitrate. Some may have 84kbits, others may have 150kbits, while others may have 50kbits but all together the total kbits will always add up to 2520kbits.

Look, when talking about the benefits of 2-pass vbr over CBR, which is inherantly single pass, we are talking about high motion scenes. In low motion scenes the CBR clip will have more bitrate than the multipass vbr clip, enough to compensate for the worse motion prediction. But during high motion scenes BOTH encoding methods will be limited to, say 2520kbits PER SEC. Within that ONE SEC the bitrate still has to be distributed PER FRAME. The multipass vbr encode has more information regarding motion and can make better allocation thus achieving better quality during high motion scenes.

I need to stress again how minimal this benefit is. We are talking about minor adjustments in bitrate at the gop level. I am not saying multipass VBR should be used if the max does not exceed your CBR level, this does not yield enough extra quality to warrant the extra encoding time. I am merely refuting the original statement that started this whole silly argument,

if you do a VBR and CBR encode and the CBR bitrate is the same as the max of the VBR, THE CBR ENCODE WILL ALWAYS GIVE BETTER RESULTS.

[LeeBear]

I'm sorry if my example was not technically correct. I just tried to keep it as simple as possible but the idea is still the same. Whether you're talking about 1 second of footage or 1 GOP structure or whatever unit you want to use. It works the same. If a 1 second of footage only requires 1700 kb to encode to maximize it's quality then a CBR setting of 2520 kbps will still use 2520 kb to encode it while a VBR would only use 1700 kb. 820 kb are just padding.

-LeeBear

[luctondo]

Hi!
I've seen it AVI files that are correctly 700mb and 715mb and i can't get my movie to the right size i want. I've tried every calculator on the net and still can't. If somebody knows the trick to get it right, please let me know!
Thank you

Hi Adam

You have listed some adavantages which maybe VBR has in this example. And you yourself admitted the quality gains are minimal. But one important point forgoten is some of the data VBR throughs away(the encoder) thinking it is not important (in order to allocate for high motion scenes) can in fact be important for subtle details in movies. This is when we are talking of bitrates in the order of 2520 & SVCD CVD resolutions.

And for Satstorm who likes practical Examples.

The mummy Reaturns, opening scenes when the scorpian king is walking accross the sands after the battle. .

Encode with CBR then Encode with VBR using the settings you gave at the beginning of this post. All encodes will look good, but now look closer at the ripples in the sand. I think the definition is better with the CBR encode.

Cheers

[adam]

D_Head that point was never forgotten, I have mentioned it numerous times in this thread. CBR will have more bitrate to work with, no arguments there. With more bitrate it can most definitely produce higher quality results than multipass vbr at a lower avg bitrate, again no arguments there. The point is that CBR will not ALWAYS surpass VBR encoding in quality, even when the VBR encode uses a lower avg bitrate and this was the only claim of yours that I ever had any contention with.

And if CBR and VBR are given the same avg bitrate and you encode the VBR in multipass vbr than it will ALWAYS be superior in quailty to the CBR clip, for reasons I have already explained. This is not something that may happen, its a mathmatical certainty. And yes this added quality will probably be minimal, but I never claimed otherwise.

So it seems you admit that CBR will not ALWAYS be superior to VBR, finally we agree so hopefully this thread will die along with your new nick. I think boathouse got the answer he was looking for a long time ago.

If you have any factual evidence to refute my statements I'll be more than willing to continue this discussion. Otherwise you are just going to have to continue talking to yourself.

D_Head -[***]- Has Entered #vcdhelp
<D_Head> hello
<D_Head> CBR VBR question
<D_Head> anyone
<D_Head> Your opinion,
<D_Head> what is better or is there a difference
<D_Head> CBR 2500 or VBR min ~1000 av 2400 max 2500, quality wise
<***> prolly the VBR
[Adam] -[****]- Has Entered #vcdhelp
[1:07:58pm] *** ChanServ sets mode: +o [Adam]
[1:08:18pm] <***> ask adam he prolyl knows more about vbr

I'm sorry but I just could not resist, that last line makes this whole thread worthwhile.

The last reply from you just goes to show your imaturity ADAM and how you can not stand to be wrong. You just cant stand that someone else has another point of view.
Discussion and debate adam I think is is what a forum is.

Personally I dont care If I am right or wrong, I learn from taking part in discussions, and as a moderater you should encourage this not just been the Know it ALL ADAM.

I have noticed on many ocations in various topics how you try to ridicule people, and put people down. (blind them with bullXXit on may occations)

And whats worse you resort to Lies to to make people look stupid

Quote ADAM

this thread will die along with your new nick. I think boathouse got the answer he was looking for a long time ago.

Because other people also agreed with the points i tried to make.. you lie and say they are my other Alias`s. Wrong ADAM.

It yould be great if TRUMAN or BALDRICK or another ADMIN would step in and confirm that I only USE 1 Alias i.e MATT=DHEAD & not JOES or any other name for that matter.

Cheers

[joes]

I think we have gragged in too many moments wich are relatively irrelevante to the initial question:

When VBR max is set to the same rate as CBR, could it ever be better in quality in any time point within the footage?

So long no one have convinced me that VBR can be better, logic tell me it will NEVER happen.

But I've never sayed CBR will ALWAYS be better, only VBR will NEVER be better. So my point is if U don't have to squeez, do CBR.

I the middle of discussion it raised a new question wich I want a qualified awnser:

How strict is the bitrate limit? In both case CBR and VBR_max? What I have been told is that Standard is abs. strict, that will mean in any 30 subsequent frames from start to the end, the data amount is not to exceed the value, i. e. frame 1-30, frame 2-31, frame 3-32,... not 1-30, 31-60, 61-90,....,that's why I sayed in 1/30 sec. resolution.

I haven't be accurate with nummers and terms, because my bad english and trying to simplify, sorry if I was misunderstood.

I don't know if D_HEAD use other alias, don't care about that neither. But everybody must have noticed that I am not a native english speaking person, actually I'm from (and in) Norway.

Maybe my question have to start a new thread, or can I get some awnsers here?

Hi Joes, ADAM is convinced we are the same person.

My last coment in the post as I have had enough of been stalked by ADAM.

Have a look here, its a little technical but some great infos....

http://bmrc.berkeley.edu/frame/research/mpeg/mpeg_overview.html

Cheers
Matt

[vitualis]

On superficial analysis, I would have agreed with D_Head.

However, if adam is correct (and I think he is) that on multipass VBR (with same max. bitrate as CBR bitrate) that the motion search precision has increased with each subsequent pass, then multipass-VBR could hypothetically look better than CBR with same max. bitrate.

Whether you could tell the difference is another matter.

Obviously, you would simply use CBR if such a situation did occur but I think that adam is putting in the hypothetical argument (and I have to say that I am convinced).

Motion search precision is quite important. For example, encode with TMPGEnc with the same bitrate CBR once on "very fast" vs. "very slow". You have the same bitrate, same material, but one looks crap and one looks good. Why?

BTW, chill out everyone. This has been a fairly good discussion.

Regards.

[NeVeRLiFt]

@D_Head

Dude you got serious issues and dont know what the hell your talking about

[DJRumpy]

I found this site very informative on the issue:

http://www.tecoltd.com/enctest/concepts.htm

Here's another

http://researchweb.watson.ibm.com/journal/rd/434/westerink.html

(These are not my words, so no flames please. Just reporting what Iv'e found. The second link is pretty interesting, as they did actual visual comparison tests with subjects, regarding CBR vs. VBR)

Conclusions
In this paper a complete system for two-pass encoding of MPEG-2 video is described. First, a pass is made over an entire video sequence while extracting relevant video characteristics. These include the number of bits for each picture, quantization scale, picture type, picture spatial activity, etc. A special preprocessing program then uses these first-pass statistics to prepare second-pass control parameters, such as the individual picture target number of bits. Finally, a second pass is made through the video sequence, but under control of the prepared parameters and a special real-time bit-production monitoring and adjustment algorithm. This second-pass control is on a picture-by-picture basis.

Owing to the limited capabilities of the first-pass CBR, some segments of the raw statistics were found not suitable for immediate use by the preprocessing. Instead, a prefiltering operation is required, including scene change and fade detection, and recalculation of the first-pass quantization scales. It has been shown that this filtering significantly aided the parameter preparation for second-pass VBR, especially in certain situations, such as immediately following scene cuts.

The general conclusion is that the second-pass VBR sequences visually appear to have a higher overall quality than the ones coded with CBR. For VBR to visually outperform CBR, a mix of "easy" scenes and "difficult" scenes is always required. If all scenes were the same (easy or difficult), the VBR results would be equal to those for CBR. The principle of VBR relies on taking bits from easy scenes and spending them on the difficult ones instead. In visual evaluations by different viewers, it was found that the visual quality of an entire video sequence is judged by the minimum quality across the whole sequence. This minimum quality is usually found in the easiest scenes, as coding artifacts are the most noticeable there. A good VBR performance is thus founded on removing bits safely from the easier scenes (i.e., without noticeably distorting their quality). These bits are then redistributed over the more difficult scenes, such that the entire video is perceived to have a constant quality.

Blah blah blah...

When it comes down to it, make your own choice, and don't worry so much about every one else's...