What is the difference between CBR-1 PASS and CBR-2 PASS?

[ChachiFace]

I thought CBR meant constant bitrate. So if that is true, then what is CBR-2 PASS? And what is the benifit?

ChachiFace

UPDATE: Nevermind...I just found the post by "Adam" on this subject.

[adam]

CBR 2 pass would mean it encodes it once in CBR, then encodes it a second time in CBR but using the information from the previous pass to hopefully improve quality.

Multipass encoding is not just for VBR. It involves much more than just distributing bitrate. Mpeg encoding is all about removing redundancy. The encoder tries to find simularities between frames in a series, and then it stores those simularities in a single frame and the following ones just add to the previous one to update the image. So even though two mpegs have the same average bitrate every second, one can still look significantly higher quality than the other because those bits are being used more effectively since the encoder was able to remove more redundant/unneccesary information. Also realize that there is no such thing as a constant bitrate. To do that you would have to encode with all I-Frames (those frames that store the redundant information) and that would completely defeat the purpose of using mpeg. CBR simply means a constant buffer rate. Its essentially still VBR but with much less variation.

Now of course the benefits of multipass encoding are going to be MUCH more obvious with VBR encoding. With CBR encoding, its probably not worth the extra time unless you are using a relatively low bitrate.

[ChachiFace]

Thanks Adam! Any chance you could answer a couple of CCE-Basic questions?

1. What is the DF option and what does it do? Does it hurt to have this checked if not needed?

2. What affect does moving the 'quality settings' slider from 'simple' to 'complex' have? Will the quality be better at 'complex'?

The user guide that came with the encoder is very 'basic' and doesn't explain much.

Thanks,

chachiface

[adam]

I use 2.5 SP and I have no option for DF. What does this stand for?

The simple vs complex slider is called image quality priority in my version, and no simply prioritizing complex will not necessarily yield better results.

To vastly oversimplify encoding, there are two general types of scenes, simple and complex. Simple, or flat, scenes contain alot of redundancy both in each frame and from each frame to the next. Examples of flat scenes are a horizon shot, a shot where the background is intentionally out of focus, or a shot with alot of solid color, usually black. These scenes generally don't need alot of bitrate and so the encoder allocates less to them so that it can free up bitrate for more "complex" scenes.

Complex scenes are the opposite. They contain alot of varying information and are usually the result of alot of movement.

So this slider lets you prioritize how CCE distributes bitrates to these scenes. Now complex scenes are always going to get more bitrate, but the problem is that you don't want to throw too much bitrate at these scenes at the expense of the flatter scenes. If you set CCE to allocate too much bitrate to complex scenes then you will get banding or contouring on flat scenes. Ever watched a VCD during a horizon shot and the picture looked hazy and lacked definition, and everything in the distance seemed to just blend? That's what you want to avoid.

On the other hand, if you prioritize the simple scenes, and you don't have an excess of bitrate, then the more complex scenes will often get the dreaded mosquito noise around moving objects. Its tiny little black dots that look like little bugs almost.

Basically you want to find the best compromise between these settings. For DVDs you should favor simple scenes, because you should have more than enough bitrate to spare for the complex scenes. For VCDs and SVCDs its really kinda a subjective decision. Some people are more opposed to one type of artifact than the other. Personally I'll take mosquito noise over banding any day. But if you use a good settings than you really shouldn't get either, but you'll just have to experiment because I believe the scale that 2.5 works on is different from the basic version. My scale is on 100 with 0 being complex and 100 being simple.

For SVCDs I usually used a setting around 22-25 depending on the complexity of the source. Now that I do DVDs I use between 30-60 depending on the source.

[ChachiFace]

Thanks for the details respnose Adam!

DF stands for dropped-frame timecode in the CEE-Basic version. That question was just answered for me in another post. I now know all about drop-frame time code which is good...I'm learning.

I guess I'll just keep my quality setting in the middle between simple and complex. Since most of my DVD's have both type scenes.

ChachiFace

[SaSi]

CBR 2 pass would mean it encodes it once in CBR, then encodes it a second time in CBR but using the information from the previous pass to hopefully improve quality.

Multipass encoding is not just for VBR. It involves much more than just distributing bitrate. Mpeg encoding is all about removing redundancy. The encoder tries to find simularities between frames in a series, and then it stores those simularities in a single frame and the following ones just add to the previous one to update the image. So even though two mpegs have the same average bitrate every second, one can still look significantly higher quality than the other because those bits are being used more effectively since the encoder was able to remove more redundant/unneccesary information. Also realize that there is no such thing as a constant bitrate. To do that you would have to encode with all I-Frames (those frames that store the redundant information) and that would completely defeat the purpose of using mpeg. CBR simply means a constant buffer rate. Its essentially still VBR but with much less variation.

Now of course the benefits of multipass encoding are going to be MUCH more obvious with VBR encoding. With CBR encoding, its probably not worth the extra time unless you are using a relatively low bitrate.

Are all MPEG encoders clever enough to do all this?

For example, an encoder doing 2-pass CBR could spend the first pass trying to identify scene changes and where to put I-frames, but Tmpgenc, for example, has a separate function to achieve this.

I believe the main "defect" in the documentation of all encoders I've tried (CCE, Tmpgenc, Mainconcept) is that they don't explain what they do.

And, theoretically speaking, an interesting variation of 2 pass encoding could be to encode first pass using a constant quality mode (user selected) and then, once scene changes and bitrate budgeting is performed, the second pass could be a VBR encode taking into account the information from the first pass. Even more interesting would be to be able to do a 3rd pass, to finetune bitrate allocation (by dropping or increasing bitrate so that the user selected average is exactly met).

But, given my understanding of how MPEG encoders work now, I doubt a 2 pass CBR would offer any visible benefit.

Your views?

[ChachiFace]

I believe the main "defect" in the documentation of all encoders I've tried (CCE, Tmpgenc, Mainconcept) is that they don't explain what they do.

I totally agree. Though I'm just a novice, I have tested a clip encoded with 1-pass CBR and 2-pass CBR and noticed through bitrate viewer that the quantization levels were much lower (which is better, right?) on the 2-pass CBR encoded clip. But, visibliy on the TV screen I could not tell the difference in quality. Though, maybe I need to view the footage with fresher eyes.

[adam]

For example, an encoder doing 2-pass CBR could spend the first pass trying to identify scene changes and where to put I-frames, but Tmpgenc, for example, has a separate function to achieve this.

It does? My understanding is that all encoders basically work the same. Its not like they can see what a frame looks like. To an encoder all they see are 16 x 16 pixel macroblocks. They compare all macroblocks to all surrounding ones in that frame, outward to a certain radius, and then they compare those macroblocks to those either preceeding or subsequent to it depending on what type of frame it is assigned to be (I, P, or B.) The numbers gathered from this test are plugged into an algorithm and from that decisions are made regarding well, everything. This information is used for locating scene changes, bitrate distribution, quantization..etc..

I assumed that when you encoded in multiple passes all this information was cached, reanalyzed and re-plugged into the algorithm. I assumed it would make all decisions of the encoder more accurate.

I haven't noticed TMPGenc working differently than any other encoder in regards to scene detection and I frame insertion. You specify the GOP structure and it will just insert I frames at a given interval. But if auto I frame insertion is enabled it will attempt to place additional I frames at scene changes. And like I said above, the information used to accomplish this is really no different than the information used for most other functions of the encoder. Are you maybe referring to the motion search priority setting? My understanding is that this simply determines how many macroblocks out it searches. Its still functioning like any other encoder, you just have the option of limiting its effectiveness in lieue of faster encoding.

I don't think multi-pass CBR encodes are nearly worth the extra time. But I do think in most cases you might notice some technical difference, as seen by ChachiFace's test (yes lower quant is better) and in some cases you might actually see a visible difference if the bitrate is low enough.

[SaSi]

Adam,

In Tmpgenc's Settings, in the GOP structure tab, you can check the "Force Picture Type setting" and then press the setting button. This will load up a form which can be used to either manualy (very tedious) or automatically scan the source video and decide where to put I-frames. You can save the list to a file. Or you can load the list from a file. For example, some demuxers can generate an I-P-B frame structure file which can be fed to Tmpenc to supposedly generate an MPEG file with exactly the same structure.

Other than that, yes, Tmpgenc will cache the first pass results (provided enough space has been alocated in Environmental settings for this), so that the second pass becomes faster.

Of the popular encoders, only CCE appears to have an algorithm that gives every new pass a chance to further improve bitrate distribution. Anyhow, it's the only encoder that does more than 2 passes.