DVD GOP

[aamir12345678]

I have a quick question about the DVD standard for GOP length. From my understanding, a GOP is a scene. It starts with an I-picture and ends right before another I-picture begins. In between are P-pictures and B-pictures. I, P, or B are all frames of the video classified as such by the encoder. The I-picture marks the scene changes, P-pictures & B-pictures are dependent on I-pictures for information (data). Now, in regards to the standard length of a DVD GOP being 18 (NTSC), what if a scene itself is (or supposed to be) longer than 18 frames? If you force the scene change because of the 18 frame limit, won't that have a negative effect on quality on the following GOP because the P-pictures & B-pictures of the following GOP are referring to a "false" I-picture for their Predictional & Bi-Directional Information (Data). Does this happen often (a scene being longer than 18 frames)? If it does, woudn't that have a significant negative effect on quality? Is this as big of a deal as I'm making it, or is it just a setting that ensures compatability at the cost of minor quaity loss?

[BJ_M]

you can have open and closed GOP's as well as variable lenth of the GOP between I frames ..

not all authoring apps except some of this but it is legit

[aamir12345678]

so you're saying that the fact that a GOP is open means that an encoder has freedom to exceed the GOP limit that you set if it deems it necessary, but it might cause problems with some authoring apps

[FOO]

scene changes shorten the GOP. A new I framed is created when the
frame is much different from the previous

[BJ_M]

here is this in action -- notice the scene change

[SLK001]

I have a quick question about the DVD standard for GOP length. From my understanding, a GOP is a scene. It starts with an I-picture and ends right before another I-picture begins. In between are P-pictures and B-pictures. I, P, or B are all frames of the video classified as such by the encoder. The I-picture marks the scene changes, P-pictures & B-pictures are dependent on I-pictures for information (data). Now, in regards to the standard length of a DVD GOP being 18 (NTSC), what if a scene itself is (or supposed to be) longer than 18 frames? If you force the scene change because of the 18 frame limit, won't that have a negative effect on quality on the following GOP because the P-pictures & B-pictures of the following GOP are referring to a "false" I-picture for their Predictional & Bi-Directional Information (Data). Does this happen often (a scene being longer than 18 frames)? If it does, woudn't that have a significant negative effect on quality? Is this as big of a deal as I'm making it, or is it just a setting that ensures compatability at the cost of minor quaity loss?

Your first miss-conception is thinking that a GOP is a scene. It is simply a Group Of Pictures. Your second was believing that 18 frames is the DVD standard for NTSC discs. 18 frames is the absolute MAXIMUM number of frames for a NTSC GOP - exceed this and your authoring program should give you a fatal error. There really is no "standard GOP" as you are thinking. Generally, most commercial DVDs use 12 frame GOPs.

It is a good thing to have a new GOP start at scene changes, but this requires multipass encoding, since the encoder has to know that a scene change is coming up. Each "scene" as you call it, can have many GOPs - it doesn't affect the quality. A closed GOP means that the "P" frames and the "B" frames cannot get any picture information outside of the present GOP - generally, there is no need to close GOPs, except at scene changes (it does add additional encoder/decoder overhead).

And "yes", getting the GOPs "right" is a big deal - but not for the concerns you are stating.

[awlchu]

BJ_M,


What tool did you use to create that GOP report?

here is this in action -- notice the scene change

[SLK001]

That's the commercial version of Bitrate Viewer. Every serious DVD creator should have a copy.

[BJ_M]

yes -- bitrate viewer commercial version

here is another shot from a different movie i made ... the GOP pattern can be all over the place in high action

[aamir12345678]

Thanks, SLK001.

[FOO]

It is a good thing to have a new GOP start at scene changes, but this requires multipass encoding, since the encoder has to know that a scene change is coming up.

I don't think I believe that. Why can't they just do a measurement of how
different the next frame is and start a new GOP. They have to be
buffered 3 deep anyway.

[BJ_M]

It is a good thing to have a new GOP start at scene changes, but this requires multipass encoding, since the encoder has to know that a scene change is coming up.

I don't think I believe that. Why can't they just do a measurement of how
different the next frame is and start a new GOP. They have to be
buffered 3 deep anyway.

FOO - you are correct , many encoders can do scene change and var. GOP on a single pass .. in fact most, if not all, of the better "pro" hardware/software encoders can do this (some can't).

[awlchu]

BJ_M,

Mixed open and closed GOPS in the same stream? I thought it was one or the other! So, when I tell my encoder (mainconcept) that I want closed GOPs, maybe I'm not...based on what? scene changes?

yes -- bitrate viewer commercial version

here is another shot from a different movie i made ... the GOP pattern can be all over the place in high action

[BJ_M]

only select closed GOPs for two reasons , 1 - your dvd authoring app requires it (there are a couple) , 2- you plan on editing the mpeg file ..


in main concept select open gops and scene detection ...

you can mix open and closed gops in the same stream .. but not all encoders can do this ..

[BJ_M]

the shot above is from procoder in mastering mode , single pass, cbr 8000

btw

[Arky]

Just as an example, one situation in which closed GOPs are compulsory is when authoring Multi-Angle Titles.


Arky ;o)

[BJ_M]

Just as an example, one situation in which closed GOPs are compulsory is when authoring Multi-Angle Titles.


Arky ;o)

mostly used only in porn i might add

[indolikaa]

Just as an example, one situation in which closed GOPs are compulsory is when authoring Multi-Angle Titles.


Arky ;o)

mostly used only in porn i might add

You beat me to the punch!

I was also going to add Sports footage...

[Arky]

Just as an example, one situation in which closed GOPs are compulsory is when authoring Multi-Angle Titles.


Arky ;o)

mostly used only in porn i might add

You beat me to the punch!

I was also going to add Sports footage...

lol

...In which case, closed curtains are also compulsory!


Arky ;o)

[dimad]

Just as an example, one situation in which closed GOPs are compulsory is when authoring Multi-Angle Titles.



Arky ;o)

mostly used only in porn i might add

And in Disney cartoons

[BJ_M]

disney makes porn ?

[indolikaa]

disney makes porn ?

You've never seen Snow White, The Miramax Years?

[flaninacupboard]

disney makes porn ?

You've never seen Snow White, The Miramax Years?

This forum never ceases to amaze me 8)

That's why we need a compendium of threads, encoder limitations of I frame insertion to childrens pornography in a few easy steps.

[duhmez]

And yes, quality could often be better if the gop COULD be longer than 18, like if it's a really long unchanging scene, there are unneeded I frames in there aing up space. However we are stuck @ 18 for ntsc just because we are.

[flaninacupboard]

that point is debatable, through a long GOP (as you find in Divx) any artifacts/errors are built up through the whole GOP, becoming very distracting. throwing in an I frame ~twice a second makes this effect minimal. i much prefer a shorter GOP and higher bitrate (in PAL land we are limited to a length of 15 anyway.)

[FOO]

Note that the GOP limit of 15 or 18 is in the DVD spec
not the mpeg2 spec , anybody that went by the mpeg2 spec
in the player firmware will play long GOPs

[SLK001]

It is a good thing to have a new GOP start at scene changes, but this requires multipass encoding, since the encoder has to know that a scene change is coming up.

Okay, to start a GOP at a scene change requires knowledge of the encoded profile of the new scene (or upcoming frames). One way to do this is with multipass encoding. Another way to do this is with buffers holding "pre-encoded" information of frames still to come. The "pre-encoded" information could be as simple as a quick, low BR encode to get the needed information on the frame to pass to the encoder.

[flaninacupboard]

well as far as i can see, all it needs is a simple "is this frame more than 50% different than the last frame? Yes, i'll make an I=frame, no, i'll follow the specified GOP." Granted that may not be possible with a realtime encoder, of the type for TV broadcast (but it should be, they have a delay of about 2 seconds on them over here!) but definitely on a PC based encoder doing CBR.

I don;t have any material to encode at the moment, if i did i'd do a quick encode and get TMPGenc to produce it's log, and see if it follows the GOP specified or can detect scene changes.

[Arky]

If you scrub along your NLE timeline, placing cuts at dramatic scene changes, you can export these shorter segments and encode them to MPEG individually. You can then concatenate them back together and 'voila', you have perfect GOP header I-frames on all your scene changes.


Arky ;o)

[vhelp]

@ Arky,

Hay, I wonder if this is at all possible inside TMPG, or as a auto/batch for TMPG
.
Or, if this were possibel to do in an .AVS script to detect, and then feed in a
"gulp" or batch to TMPG or something. Maybe sounds stupid, but I'd be interested
to know if it were at all possible - - I don't know. Anyways..

-vhelp