CBR vs VBR with same file size.

[PuzZLeR]

I honestly could not find my answer in searches on this.

Let's say you encode/render one source clip to MPEG-2 in three different ways with the same encoder:

1. VBR
2. VBR 2-pass
3. CBR

Hypothetically speaking, let's just say that you had the initial encoder settings a certain way, and you ended up with the exact same video bitrate/file size for each in the final result. The only different settings are different min/max/avg bitrate entries. All other things are the same: filters, effects, etc.

Would one be better quality than the other (if at all)? I'm talking motion scenes, pixelation, blockiness, etc. I'm just wondering if one method is more optimal given the same resulting bitrate.

The answer would be obvious if there's a difference in bitrates, but keep in mind, and I must repeat, we have three clips with the exact same file size/video bitrate, but all three got there differently. The question asks if there's anything in the encoding/rendering that gives it an extra edge, albiet small, or is it as simple as quality is only a function of bitrate regardless of how you got there.

Thanks in advance to the bright minds here!

[wulf109]

VBR-2Pass

[Cornucopia]

If you're saying AVERAGE bitrate is the same for all, then I'd say 2passVBR (or more) is almost always going to be better than CBR. 1passVBR could vary between being better than CBR, to worse than CBR (but never quite as good as 2pass).

If you're saying MIN=AVG=MAX=same for all 3 types, then I'd say there's probably NO difference. (Might be a slight difference/improvement with 2pass because of greater accuracy in Motion Estimation).

Scott

[guns1inger]

It depends on the content of the clips, and the bitrates used.

If the bitrate is high (CBR and avg for VBR) then there may be little difference if any.

As the bitrate drops down, especially below the 7500 - 8000 mark, 2-pass VBR starts to get a definite advantage.

It isn't just the average bitrate that is important, but how the bitrate is allocated across the video. CBR allocates the same bitrate throughout, regardless of the content of the scene. VBR allocates bitrate based on need, while working around the limitations you have set it with the min/avg/max bitrates.

You could still get the same file size and the same average bitrate, but get very different quality during certain scenes. For example, your video clip is a mixture of newreader (talking heads, mostly static) and crosses to the story being covered, which is a fire (smoke, flame, dark, lots of movement). The CBR encode will allocate the same bitrate to the head shots as it will to the action, even though the static shots don't really require it. The smoky action, on the other hand, needs a higher bitrate that CBR cannot give it.

Contrast this to a 2-pass VBR, where the head shots get a much lower, though still adequate, bitrate, allowing the encoder to give the action shots a higher bitrate when required.

The two files have the same average bitrate and file size, but the CBR version suffers from artifacts during the action scenes that the VBR version doesn't have.

This is a just a simple example, and as such, is only representative of a small range of scenarios.

The real moral of the story is that there is no hard and fast rule. The needs, and results, vary based on the content of the video and the bitrates you have to play with.

[PuzZLeR]

I'm grateful for your incredibly fast and very informative respones. Thank you kindly.

So, in my understanding, VBR 2 pass "intelligently" better allocates bits whereas CBR, in rather high probability, will either "lack" bits, meaning less bits for higher motion, or "pad" bits in many still scenes that don't need it too much.

Ok, clear as clear.

But just for the sake of better understanding, this leads me to believe, at least theoretically, that if all three methods resulted in an avg bitrate, such as the mentioned 7500 - 8000 kbps, there would be little difference.

If I understand correctly, is this because of a DvD ceiling?

So, due to this very ceiling/max there would be a smaller margin for moving info around, which would stifle data distribution somewhat. Therefore, less variance among the three clips.

So this rate of difference in quality decreases when near a limit of some sort.

This logic also applies if you narrow your min/max bitrate parameters at the beginning. Less margin for movement there too, hence similar quality in all three methods then.

Ok, I think I'm understanding it...

[45tripp]

VBR is still more bit efficient.
It's also dependent on how good your encoder's vbr mode is.

If you were approaching the ceiling and def wanted to utilise max bitrate for the most demanding scenes, you'd probably be better off using cbr to ensure max bit allocation and not worry about bit wastage in less demanding scenes.

[guns1inger]

The "DVD ceiling" is in fact 9800 kbps for video, with a combined bitrate of 108000 kbps. Unfortunately, given the variable quality of some players and media, not everyone can play back bitrates that high successfully. Even so, high 8000's is not unreasonable.

As for VBR vs CBR - if you are using a very narrow range for your variable bitrate parameters then you are basically doing a long winded CBR encoding and not getting the benefits that VBR provides. To me this says you really don't understand the difference, or when to use what.

If you use are using a narrow range with a high average then you should be encoding CBR instead.

If you are encoding with a narrow range at a low bitrate or using CBR at a low bitrate (assuming full-D1 resolution) then you should be using a wider range VBR encode.

A good encoder will hit it's average to within a few kbps on a two pass VBR encode, so you know you will hit your file size target. Therefore there is no reason to use narrow range parameters for VBR encoding.

[jman98]

Guns1inger provided some really good info. I'll give two practical examples.
Case one has about 2 hours of video you want to put on a single layer DVD. Let's say a bit rate calculator says you can use a bit rate of 4000 Kbps for the video. One option would be to do VBR encoding with an average of 4000, a maximum value of 6000 and a minimum value of 2000. You could also do a CBR encode at 4000. You should expect the VBR (especially if 2 pass) to be better because the encoder may go as high as 6000 for complex parts of the video and the CBR encode will never go above 4000 for the same segment. VBR will use a lower bit rate where it makes sense. For example, if you have a small segment that's all black, it makes no sense to waste bit rate on an all black scene, so the encoder will render this at 2000 or close to it as in this example.

Let's say you have 60 minutes of video and you can use a bit rate of 8000. You could use VBR with an average bit rate of 8000, a maximum of 9000 and a minimum of 7000. You could also use CBR encoding at 8000. You won't see much, if any, difference at bit rates this high between VBR and CBR encoding.

VBR encoding is used at lower (ie. below 6000) bit rates to squeeze as much quality as possible out of the video by allowing the encoder to exceed the average bit rate when it makes sense to use lower than average bit rate (to save space) when that makes sense. CBR just gives a constant quality performance across the board, but for complex scenes as in my first example, a CBR encode at 4000 may look very inferior to a VBR encode that uses 6000 for the same scene but has an average value of 4000. The CBR encode is wasting bits in my example by encoding the solid black scenes at 4000 when it's not necessary as a lower bit rate would not look any worse to most people. VBR uses the bits in a more intelligent manner whereas CBR encoding is just "one size fits all" for everything.

[AlanHK]

I generally use HCEnc, which is a 2-pass VBR encoder. If I leave its window open with preview on, I can see the rate for each scene as it encodes.

First it goes through the entire video, analysing it. Then it begins its second pass, creating the MPEG, and using the stats it collected, it knows how much it can allocate for each scene. I can see the rate go way above the average for busy sceens, and way below for dark scenes with little movement.

Sometimes I use DivxtoDVD, which is a single-pass, CBR encoder. It takes less than half the time than HCEnc, but in an hour I can expect it to have two or three scenes broken up by pixellation. The files it makes for similar quality are larger than HCEnc. But I use it if I just want to watch something on my player quickly and delete it.

[edDV]

CBR wins in two cases.

1. Many communication channels (not DVD) require constant bit rates and reject VBR.

2. At higher DVD bit rates VBR and CBR converge on the same thing as the spread between maximum and average VBR bit rates narrow. Higher bit rates are recommended for noisy or shaky hand held camcorder material. VBR can actually produce inferior results if motion estimates are faulty or if the VBR analysis over compromises the low motion end to meet the average bit rate goal. Slow moving darker areas could become blocky.

[Alex_ander]

CBR wins in two cases.

I'd add one more case: CBR also wins when you can't afford multiple-pass for timing reasons (and still want predictable file size). It's when you have to use a slow Avisynth filter (e.g. FFT3DFilter); then even in CCE each pass can take 5 times more than actual film duration.

[guns1inger]

It's when you have to use a slow Avisynth filter (e.g. FFT3DFilter); then even in CCE each pass can take 5 times more than actual film duration.

Buy more disc and write to an intermediate file using a lossless codec. I have a couple of nice noise removal scripts for VHS transfers, however they average around 11 - 12 hours per pass to process a 2 hour video. I do a single pass to a Lagarith encoded file, then encode to mpeg2 from there. Cuts down a 30 - 40 hour encode to around 15 - 16 hours, and lets me use multi-pass VBR encoding.

[PuzZLeR]

I thank you all for your posts. It was most helpful.

Here's my understanding, assuming the final file sizes are the same:

CBR = VBR when the entered max bitrate(for VBR) = min bitrate(for VBR) (in theory).

The shorter the range between max bitrate(for VBR) and min bitrate(for VBR) the more closer the quality it will be to CBR.

If you're encoding at the assumed maximum bitrate allowed, then CBR = VBR. You'd be better off with CBR then for the speed.

At lower bitrates, the motion scenes will look nicer in VBR, but maybe, just maybe, some of the still scenes will look nicer in CBR (since we may have only so much bitrate to work with, and an equal amount to allocate between the two schemes).

Due to the advantage of "foresight", 2 pass VBR is higher quality than 1 pass VBR because 2 pass allocates bits better to the scenes that need it most, and will "pad" less to, for example, black screens and such. Likewise, this efficiency and padding would be even more obvious still with CBR.

2 pass CBR (if it exists) = 1 pass CBR in quality. In other words, the 2nd pass on CBR is redundant, and in theory is pointless.

[edDV]

I thank you all for your posts. It was most helpful.

Here's my understanding, assuming the final file sizes are the same:

CBR = VBR when the entered max bitrate(for VBR) = min bitrate(for VBR) (in theory).

The shorter the range between max bitrate(for VBR) and min bitrate(for VBR) the more closer the quality it will be to CBR.

If you're encoding at the assumed maximum bitrate allowed, then CBR = VBR. You'd be better off with CBR then for the speed.

At lower bitrates, the motion scenes will look nicer in VBR, but maybe, just maybe, some of the still scenes will look nicer in CBR (since we may have only so much bitrate to work with, and an equal amount to allocate between the two schemes).

Due to the advantage of "foresight", 2 pass VBR is higher quality than 1 pass VBR because 2 pass allocates bits better to the scenes that need it most, and will "pad" less to, for example, black screens and such. Likewise, this efficiency and padding would be even more obvious still with CBR.

2 pass CBR (if it exists) = 1 pass CBR in quality. In other words, the 2nd pass on CBR is redundant, and in theory is pointless.

One more issue for some encoders.

Two pass VBR can be set to exactly fit a target file size (e.g. DVD-5 or DVD-9). CBR should also get exact results based on constant bit rate multiplied by seconds. Single pass can get file size errors that exceed DVD capacity. Common practice is to allow some safety margin for single pass or live VBR file size error.