VBR vs CQ

[Shadowmistress]

Why do most people on this forum prefer VBR over CQ? You have the same amount of control over each of them and they do the exact same thing but CQ only requires one pass.

VBR:
Average Biterate 1150
Maximum Biterate 3000
Minimum Biterate 300
2 pass

CQ:
Maximum Biterate 3000
Minimum Biterate 300
Quality 50% (Will give you 1500)
1 pass

Quality determines the average biterate setting.
[Maximum x Quality=Average Biterate]

[3000x50%=1500]
[2000x50%=1000]
[1000x50%=500]
[3000x75%=2250]
[2000x75%=1500]
[1000x75%=750]
[3000x25%=750]
[2000x25%=500]
[1000x25%=250]

OR
[Preferred Biterate/Maximum=Quality]
[500/3000=0.166 or 16%=480kbps, 17%=510kbps]

See my point?

[BJ_M]

often the Q is higher in CQ than in multipass VBR

[ZippyP.]

Firstly, CQ is VBR.

2-pass VBR can better distribute bitrate because it has done an analyzing pass so it knows the amount of high action vs. low action scenes while CQ can only do a best guess. 2-pass will hit the file size very closely while CQ cannot.

[sacajaweeda]

Firstly, CQ is VBR.

2-pass VBR can better distribute bitrate because it has done an analyzing pass so it knows the amount of high action vs. low action scenes while CQ can only do a best guess. 2-pass will hit the file size very closely while CQ cannot.

What he said.

[bugster]

Why do most people on this forum prefer VBR over CQ? You have the same amount of control over each of them and they do the exact same thing but CQ only requires one pass.

VBR:
Average Biterate 1150Maximum Biterate 3000
Minimum Biterate 300
2 pass

CQ:
Maximum Biterate 3000
Minimum Biterate 300
Quality 50% (Will give you 1500)
1 pass

CQ does NOT allow you to set an average bitrate, so you cannot accuratley determine the final filesize. For some people/circumstances this is important.

Quality determines the average biterate setting.
[Maximum x Quality=Average Biterate]

[3000x50%=1500]
[2000x50%=1000]
[1000x50%=500]
[3000x75%=2250]
[2000x75%=1500]
[1000x75%=750]
[3000x25%=750]
[2000x25%=500]
[1000x25%=250]

OR
[Preferred Biterate/Maximum=Quality]
[500/3000=0.166 or 16%=480kbps, 17%=510kbps]

Are you certain this is true. If so then why have some people (and I am thinking of Kwag (www.kvcd.net) and his followers) spent a lot of effort in developing complex tools and procedures for determing the correct CQ settings required for the desired filesize.

I agree that the quality of CQ is excellent and the encode time is much less than 2-pass VBR, but these are not always the overriding factors of determining encode mode. I do use CQ often, but when the final filesize is importnat, I am happy to use 2-pass VBR and wait the extra time.

[Shadowmistress]

Kwag and his followers must be smoking something. Check out my trials:

Source:
Type: avi
Length: 3:02 min
Frames: 25 fps
Size: 320x240
Biterate: 54/kbps
Sound: 48000 hz stereo Mpeg Layer 3-3,63/kbps
Size: 9,858 kb

Conversion:
320x240
23.976 fps
lowest quality (very fast)
48000 hz stereo at 64/kbps
Type: MPEG-1 System VBR

Stills captured during motion (camera pan from side to side).


Again, why do people prefer VBR?

[pijetro]

I think you're way over your head here....
Your VBR method is wrong..

1) 320x240 encodes is not what you should be judging with..
2) A VBR needs a CBR pass to be able to encode. So in essence, a VBR is a CBR with bitrate allocated more efficiently..
3) You've created a CQ, but haven't told us the average bitrate, so that you can compare to the VBR..
4) Where the heck are you getting your minimum and maximum values from??
How can you possibly give a minimum of 100, and an average of 200??
The reason why your CQ encode looks better, is that CCE decided to jam the bitrate to max, in order to keep good quality...

Again, why do people prefer VBR?

'Cause they've got a required filesize, and they know how to use it properly...

[bugster]

Kwag and his followers must be smoking something.

You may well be right there

As to your samples, I can't really comment as I never encode anything with that low a resolution or bitrate. Maybe CQ is MUCH better than VBR at those kinds of numbers. When you start talking about DVD resolutions and bitrates (and mpeg-2) then VBR and CQ are close enough in quality that I personally have trouble telling the difference. Again, I use VBR when accurate filesize is more important than time.

[Shadowmistress]

I think you're way over your head here....
Your VBR method is wrong..

1) 320x240 encodes is not what you should be judging with..
2) A VBR needs a CBR pass to be able to encode. So in essence, a VBR is a CBR with bitrate allocated more efficiently..
3) You've created a CQ, but haven't told us the average bitrate, so that you can compare to the VBR..
4) Where the heck are you getting your minimum and maximum values from??
How can you possibly give a minimum of 100, and an average of 200??
The reason why your CQ encode looks better, is that CCE decided to jam the bitrate to max, in order to keep good quality...

Again, why do people prefer VBR?

'Cause they've got a required filesize, and they know how to use it properly...

1) Don't like the settings? Do your own trial and post the results. This was just an example. I purposely encoded low to make the difference more noticeable.
2) My point is that CQ does the same thing as VBR
3) CQ average biterate is (maximum biterate x quality % = average biterate) Read the first post again. (1000 x 20% = 200 kbps)
4) I chose weird max/min/ave biterates to illustrate the control one may have. If CCE did what you say, why is the filesize smaller in CQ?

I invite people to do trials on 3 minute clips and post the results. Try sticking your ave/max/min settings in CQ (using my formula) rather than VBR and see if you don't get better quality AND smaller filesize.

[FulciLives]

A 2-pass or multipass VBR encode gives very "odd" results when you are encoding a small sample ... and a clip of 3min 02sec is a small sample.

You cannot expect a 2-pass or multipass VBR to work correctly that way.

So your test results, at least in this instance, cannot be trusted.

- John "FulciLives" Coleman

[pijetro]

Shadowmistress...

Stick every one of your four clips into Bitrate viewer, and tell us what the average bitrate, and Q level are please...

[junkmalle]

Shadowmistress...

Stick every one of your four clips into Bitrate viewer, and tell us what the average bitrate, and Q level are please...

You can see from the file sizes that they're all close to the same average bitrate.

I use Xvid most of the time and I stopped using two pass variable bitrates too. I use Xvid's quantization mode -- essentially a single pass constant quality setting. I don't know beforehand exactly how large the file will turn out but I know that every frame will have the desired quality.

I usually encode 640x480, 23.976 fps, Xvid quantization value 3, and the final bitrate turns out anywhere from about 800 to 2500 kbps, usually around 1200 to 1500. The results are almost as good as the original source (usually digital cable caps).

[johns0]

It is impossible to accurately calculate the average bitrate with cq,i can do a wide screen 352x480(23.976fps) at 5000kbps max with 3000kbps average and 300kbpsmin and the resulting bitrate can be anywhere from 900 to 2000 kbps at 80%cq,full screen 352x480(29.97ntsc) can get up to 3000kbps at 80%cq.

[pijetro]

Johns0 wrote

resulting bitrate can be anywhere from 900 to 2000 kbps at 80%cq

How is there a percentage derived from CQ??
I thought that the lower the number (0-40,40-60,60+), the better the quality??

BTW, CCE has a great .PDF file at the very bottom, including charts, explaining the different quality levels , and how bitrate gets allocated to specific modes..

[lordsmurf]

Shadowmistress, for low bitrate MPEG VBR, you're using a pretty crappy encoder. I can already guess you're using TMPGENC. Figures. It was made for CQ. End of story.

If you want to see a good encoder in action, give PROCODER a spin. Put it on 2-pass mastering mode.

[adam]

Ok, a couple of things that some others have already stated. CQ IS VBR, it just uses a 1pass quality based algorithm to allocate bitrate. You choose a quantization level and it adjusts bitrate to reach this. Naturally it is impossible to come up with any kind of formula for predicting filesize or average bitrate across the board. You have to test each source individually for compressibility because it can take drastically different amounts of bitrate to get two sources to have the same quantization level.

Multipass VBR on the other hand lets you set the average bitrate and it instead adjusts quantization per scene to reach this level.

Now for the differences. When using CQ mode any given scene may have a lower, and thus better, quantization value then when encoding with multipass VBR, but multipass VBR should always have a lower average Q value then a file encoded in CQ at the same average bitrate. This is better. In my opinion multipass VBR is a superior method of encoding.

Now as for your sample encodes, first let me say that your clips are too short to be meaningful. The whole point of VBR, and especially multipass VBR, is to allow fluctuation of bitrate according to the complexity of any given scene. With 4 mins or so, out of what looks to be a Survivor episode, I doubt there is much variation in your source. I don't think variable bitrate encoding is really gaining you much with these samples.

Secondly, its pretty much impossible to draw comparisons between single frames (IMO) and its most certainly impossible if you don't compare identical frames encoded using identical pictures. An I frame is an entire picture, and will have many more times the amount of bitrate then a P or B frame, which only contains information regarding the non-redundant data in a scene. I can take an encode at CBR 9mbit and find many P frames that look substantially worse then an I frame from a 1mbit encode.

Now for a practical explanation of why the CQ looks so much better....the answer is that it probably doesn't. A laymens explanation of quantization is that it is the amount of data that the encoder throws away. So obviously the lower the Q value the better. To achieve so little bitrate in your quality based encode, I assume your quantization level was very high. Thus alot of data was thrown out. As for your multipass and CBR encodes, you allowed the encoder to adjust the quantization level according to your bitrate. I bet if you analyze your encodes you will see that these encoders have a much lower quantization value. This means that more of the source's information was retained. This explains the extreme macroblocking that isn't as present in the CQ encode. The problem is that all these encodes look like total shit. They all look like mush so there's no way to compare the detail in the scene, all you can do is count the visible macroblocks. If you used a higher bitrate then there would be less macroblocking and my guess is that the 2-pass encodes would look much more detailed. I could easily duplicate your CQ results in multipass encoding by manually raising the Q (I'd have to use another encoder with more control though) or I could run it through a noise reduction filter first.