Keyframes, what's best and what's acceptable?

[boombastic]

I read an article about keyframes. Also rede a article on digitalfaq.com about capturing and it says a little about keyframes using PowerVCR.

I know the basics about keyframes:
I-frame - Intra frames
B-frame - Backward frames
P-frame - Predicted frames

So, when capturing, you must stabilish a number for B-frames and P-frames since the I-frame is always the whole frame and uses a lot of space, being needed only one to give directions to the following and previous frames and a couple of seconds. B-frames the frames that came before the next I-frame and P-frames the ones which come after. For what I understood is that these settings affect A LOT on how macroblocks will or not appear that much on your capture (please correct me if I'm wrong)

On Digitalfaq.com guide on capturing MPEG with other card it says (for the options on PowerVCR II):

The IPB settings are optimal at 2B/4P or 2B/3P but can be 1B/2P for less demand on CPU. Never use I-Frame only as that is not a DVD-spec MPEG file.

Can somebody explain these settings a little bit more or explain why a certain setting is best for quality or CPU?

[SaSi]

Your question is joining together two different - and conflicting - issues:
Capturing and Encoding. So my response has to be a bit technical and complicated.

Capturing (with a software codec) requires settings that provide:
optimum performance (low CPU load so that CPU isn't chocked and frames are not dropped),
moderate file size (HDs are cheap nowadays) ,
lossess encoding (as much as possible so that we don't lose even more)
and
high editability (typically you need to trim and otherwise edit the captured material).

Encoding is a process where you typically take finished material and encode it in such a way that:
it fits in a given space (e.g. a DVD),
destroys as little as possible from the original quality
and
conforms to the requirements of the media (the streams must have a precice - more or less - syntax)

Now, each of the two tasks above are demanding in their own right. Trying to combine them in a single stage (capture to a video stream that can be directly authored to a DVD), is sometimes too much.

Not that it's not possible.

(The above was an introduction for a better understanding of some simple facts).

A video file can be compressed badly, good and better. Even with the same encoder program. Different settings provide better quality and require more CPU power - therefore - time. All that with the same file size as an output file.

When capturing, you only have 1/25th of a second to capture and encode a video frame (in PAL).

There is a limit on what can be done in such a short time.

To capture and encode in real time, certain shortcuts must be made (now talking specifically about MPEG-2 capture for DVD authoring).

I frames are pretty fast to encode but take up much space.
P frames are fast to encode, and take up much less space (about 1/5th of an I-Frame as I found).
B frames take the most time to encode - arguably twice as much as a P frame - and take up less space than a P frame (my experiments how 1/2 ~ 1/3rd of a P frame).

So, the easy conclusion is that the more B frames you use, the slower the encoding will get. Depending on your CPU power, it may or may not be able to catch up with real time encoding when you use many B frames between P frames.

It's possible to use a sequence like IBBPBBPBBPBB... where you get two B frames between two P frames.
Or IBBBPBBBPBBBP... where you get three.
Or IBBBBPBBBBPBBBBP... where you get four and even more time to encode.

As each B frame requires less space, video encoded with 4 B frames in a row requires less space for storing.

When combining the capture process with the encoding process, then you are also "fighting" the given space limitations - you must limit the bitrate so that the video will fit on a DVD.

Therefore, if you are working with a constant bitrate, more B frames mean more pixels will be "salvaged" from the video and you will get better quality.

Imagine macroblocks as huge pixels where the video contains a single value for every group of 8x8 pixels. Something like reducing resolution from 720x576 down to 90x72. Such a video looks terrible.

So, B frames help in avoiding the macroblocking "feature" without having the filesize skyrocket.

Now, if the above help you understand the issues involved, you may end up with your own conclusions - which is the best way.

If more questions come up, ask them.

[boombastic]

Ok. First, thanks a lot for taking your time to answer my question.
Now, let's see if I understood (let's take digitalfaq.com's settings for example).

Question 1
If the Setting is: 2B/4P, that means: IBPPPPBPPPP ?
If the Setting is: 1B/2P, that means: IBPP ?

Question 2:
The higher the B frames used, the better the quality and less space, but it takes a good CPU to catch up?

Question 3: (continuing from Question 2)
So I guess the only way to know what is the optimal for me is to try lots of settings and see which one gave me no dropped frames with the most quality?

Question 4:
What is the difference between having 1B/2P or 2B/4P (cause mathematically they are the same)? Would it be more processing time (CPU usage) but more details?

Question 5:
Maybe the guy who wrote the article on digitalfaq.com might be the better one to answer this but, why on the settings he suggested he always said to use more P frames than B frames?

Question 6:
What is your optimal suggestion for keyframes capturing in real time? I have a P4 1.5 ghz 512 RAM, Maxtor HD 7200 RPM.

Thanks again for your attention.