Is IBPBPBPBPB the best all-purpose GOP structure?

[Ronaldus]

Having gotten through most of the hurdles of making DVDs from captured material, I'm trying to improve the encoding step. I used the guide at
http://dvd-hq.info/Compression.html for most of my settings.

- The guide at dvd-hq.info suggests: IBPBPBPBPB (1,4,1)
- TMPGEnc's "Standard" default is: IBBPBBPBBPBBPBBPBB (1,5,2)
- I've also seen a few other variations like IBBPBBPBBPBBPBB (1,4,2)
and (1,5823,2) <too long to write>

Could those who understand the meaning of these GOP structures or who have tried these common formats please tell me which is the best all-purpose structure to use?

I'd like to find something that works well for all possible video types or at least some sort of fixed rules such as use GOP structure X for animation, structure Y for home video, structure Z for TV captures, etc.

My goal is to have 1 (or more) .mcf files for TMPGEnc which I can select based on well-defined criteria (not something vague like "the video has fast motion") so I will only need to change the average bit rate based on the length of the video I'm encoding (I pick 4000 min, 8000 max). Since quality is more important than encoding time, I always use the highest quality settings with 2-pass VBR which ends up about a 9 to 1 encoding time ratio on my system.

Do people agree for those who are not "tweakers" it should be possible to use the exact same TMPGEnc settings (other than bit rate) on all types of video input and get good results as long as the average bit rate stays well above 5000 Kbps? In my opinion "good results" means very rare/very little blockiness and the MPEG-2 output looks almost as good as the source material most of the time.

[SatStorm]

IBBPBBPBBPBBPBB offers the best compatibility of all

About quality, it depends... Anyone has his suggestions on this.
Personally, I believe that with DVD bitrates, it doesn't really count that much

[SLK001]

Most commercial DVDs use: IBBPBBPBBPBB. The maximum GOP size for NTSC is 18. For PAL, it is 15. Some authoring programs give you problems when you present them with GOPs at the limit.

I use the commercial GOP of IBBPBBPBBPBB for most of my encodes.

[ronnylov]

A B-frame is more compressed than a P-frame which is more compressed than an I-frame. The I-frame is just like an jpg-picture, the P-frame is based on differencies from previous frame and a B-frame is calculated by motion vectors, parts of the picture that have moved since previous frame.

If you want high compression it may be better to use many B and P frames and a long interval between I-frames. But if the GOP is too long the picture is degraded too much before being refreshed by a new I-frame. On DVD the GOP lenght is limited to max 15 with PAL and max 18 with NTSC. If the bitrate is very high the quality may be maxed out and using less B-frames and less P-frames may get a better picture.

In most cases the IBBPBBPBBPBB GOP structure seems to be a good compromise at DVD bitrates.

If you want to tweak you can try other settings as long as you don't use GOP's longer than DVD specification allows. If you want more tweaking you may want to try different quantizer matrices.When encoding interlaced miniDV at high DVD bitrates I think using the "mb1 interlaced DV" matrices gave me much more improvement than any other tweaks. Too bad that dvd-hq.info just say not to touch the matrices... And when encoding progressive material you need another type of matrices. If you want the best quality with tmpgenc you must optimize the settings for each type of source file. But if you're not a tweaker then just leave it at default and accept the results.

You may be interested in trying another encoder like Canopus Procoder or Mainconcept encoder, which in my opinion, both give better results than TMPGEnc. Canopus Procoder is easier to set up than mainconcept encoder and may be a good choice if you don't want to tweak too much. TMPGEnc mostly gives a sharper picture but often suffers from blockiness. Canopus Procoder is a little bit softer but in most cases cleaner with less artifacts in the picture. Mainconcept encoder is somewhere between but more tricky to set up than TMPGEnc but it also allow more fine tuning if you are a tweaker...

[Ronaldus]

Sorry for the delay following up to the 3 good replies. The extra details you included were especially helpful, Ronnylov.

To summarize the comments:

- Ronnylov and SLK001 both recommend: IBBPBBPBBPBB which is also what commercial DVDs use (that is a BIG endorsement for this one)
- SatStorm recommends: IBBPBBPBBPBBPBB
- The dvd-hq.info TMPGEnc tweaking guide recommends: IBPBPBPBPB
- TMPGEnc's "Standard" default is: IBBPBBPBBPBBPBBPBB

Based on Ronnylov's comments, I can understand that the more frames between I frames, the worse the picture quality, so that would seem to argue strongly against the longer TMPGEnc's Standard default and slightly against SatStorm's structure which is a little longer than the other 2.

The only difference between what commercial DVDs use and the dvd-hq.info guide I've been using up to this point is 1 vs. 2 adjacent B frames. If B frames are motion vectors, then doesn't having 2 rather than 1 adjacent B frame imply slightly lower picture quality?

Ronnylov said that TMPGEnc tends to be sharper (which is important to me), but sometimes also gives blockiness. I've encoded several things using IBPBPBPBPB in TMPGEnc and I have only seen blockiness on very rare occasion (typically on transitions into a new scene). I guess I haven't seen it enough (or I don't know where to look) to be turned off on TMPGEnc... yet.

Now that I think I have it narrowed to 2 possible GOP structures as the best "all purpose" one for reasonably good quality (obviously all I frames would be the best), can people throw in some tie breaking opinions for:

IBBPBBPBBPBB vs. IBPBPBPBPB?

Logically, IBPBPBPBPB should have better quality due to fewer B frames and less frames between I frames, but the other structure is the "commercial" standard and I assume there was a good reason the standards committee picked the one with 2 B frames over the other?

[Perro Grande]

I'm sure I'm committing a cardinal sin related to compatibility, but when I'm serious about quality, I set TMPG to use 1 I frame, 4 P frames, and 0 B frames.

This certainly compromises compression, too. However, to my eye, the results are noticeably better than TMPG's standard GOP.

When I'm willing to sacrifice some quality for compression, I use the most common commercial DVD version: IBBPBBPBBPBB. However, I'm going to give IBPBPBPBPB a try to see if I can notice a difference.

[ronnylov]

Using a longer GOP with more B-frames can give better quality at low bitrates, because it compresses more effeciently. The "suggested standard GOP" of IBBPBBPBBPBB seems to be a good compromise at DVD bitrates (at least for me when encoding PAL video). I think you probably need much higher than DVD maximum bitrate to use I-frames only. If you have a bitrate at 8 mbit/s or higher you can try a GOP with less B-frames, like IPPP or IBPBPBPB. When you remove B-frames you get less bits over for the I-frames and P-frames (at the same bitrate) so the average quality may actually suffer. It is important that the quality of the I-frames are good because the other frames are based upon the I-frames. The results may vary with different type of source video so you have to test and see what looks best in your case.

In my last post I said I-frames are less compressed than P-frames and B-frames. But this also means that an I-frame need more bits than the other frame types to look good. When the bitrate is limited you have to find a compromise to make everything look fine.

[lordsmurf]

I use 1-2-2 and 1-3-2 on live MPEG-2 captures from ATI.
I use 1-4-2 on other, which is max 15-length GOP.
1-5-2 is the max for a 18-length GOP.

The 1-5000-2 thing is just the retarded KVCD template, which is made up of doing many things the wrong way. It tops out at the GOP-max anyway, never hit 5000.

Quality is determined by movement. But more "I" means bigger file size, though not always that noticeable.

[SLK001]

All the GOP sequences mentioned here are DVD compliant EXCEPT the one used by Perro Grande. A DVD stream cannot be of the "low delay" type, so must contain "B" frames.

[Perro Grande]

I thought that a "low delay" stream was made up of ONLY I frames.

Obviously, I've only been able to test on a fairly limited set of DVD players, but every single one I've tried has been able to play the IPPPP GOP. Obviously, I'm not going to sit here and proclaim it "compatible," but I am under the impression that the DVD spec allows for a stream without B frames. I know that it does not allow an I-frame only GOP.

[lordsmurf]

I thought that a "low delay" stream was made up of ONLY I frames.

Obviously, I've only been able to test on a fairly limited set of DVD players, but every single one I've tried has been able to play the IPPPP GOP. Obviously, I'm not going to sit here and proclaim it "compatible," but I am under the impression that the DVD spec allows for a stream without B frames. I know that it does not allow an I-frame only GOP.

Ask BJ_M about this, honestly.

[SaSi]

I love this "under the hood" kind of subject. I've also read some pretty interesting point of views for the GOP structure (one of my earliest nightmares).

The IBBPBB... GOP structure just "HAS" to be the best general purpose structure as it is quoted in every MPEG2 description around. So, I doubt there is even one MPEG2 decoder out there unable to play that back.

I've played with GOP variations more than I've actually encoded material. And this is what I've found:

Although more B frames replacing P frames will improve quality at a given bitrate, one has to drop the bitrate too low to actually notice the improvement. And with low, I mean 2300~2500kbps for a 720x576 25fps stream.

What I've seen makes the actual difference is how clever the encoder is to break the GOP sequence and insert an I-frame when there is significant scene change.

Almost all encoders I've tried (including Tmpgenc, MainConcept, Canopus, Honestech, Hero, ... ...) will almost religiously follow the IBBPBBPBB... structure and even try to encode scene cuts within B frames.

Fortunatelly, Tmpgenc has an option to scan the stream and look for scene changes altering the GOP structure and GOP lengths starting new GOPs at scene changes. However, it takes a lot of time to scan (and even longer to encode), so I guess few people are using this.

CCE is the only encoder that doesn't fell into the most common traps. In CCE you must deliberately choose GOPs to have equal size for CCE to ignore scene changes.

What I mean, with all this, is that the best GOP structure is a variable GOP length one, decided by the encoder so that drastic frame differences are encoded into an I-frame.

[SLK001]

What I mean, with all this, is that the best GOP structure is a variable GOP length one, decided by the encoder so that drastic frame differences are encoded into an I-frame.

I agree with SaSi here. A CCE VBR encode is the best that I have seen. It takes more time, but the results are worth it.

[duhmez]

A VBR CCE encode taking more time???!??!

more time than what? CE is lightning fast....

[Ronaldus]

Let's see if I understand... too few B frames doesn't allow the overall compression to be as high as it could be which may deprive I frames of better quality to stay within the bit rate limits, but I think SaSi said that additional B frames won't make a difference except at low bit rates?

Perhaps there are bit rate ranges when IBBPBBPBBPBB is preferable over IBPBPBPBPB? FWIW, the stuff I encode usually requires an average bit rate of 5600 to 7800 Kbps.

There has been enough praise for the (1,5,2) commercial standard with 2 adjacent B frames to convince me to switch to it for now. Perhaps it will improve the slight blockiness I see on scene transitions (a good example would be the opening fade-in on the Simpsons). I was using the IBPBPBPBPB (1,4,1) structure because dvd-hq.info's guide recommended that. It would be interesting to hear why the author of that guide chose 1,4,1 over 1,4,2 or 1,5,2.

SaSi, when you said most encoders (except CCE) "will almost religiously follow the IBBPBBPBB... structure and even try to encode scene cuts within B frames", are you saying if a scene change lands in a B frame that the encoder will attempt to do it by motion vectors (if that's the right words) rather than inserting an I frame at the scene change? If this is true, then I'm puzzled what good is the "Detect scene change" option in TMPGEnc (or any other encoder besides CCE)? I can see where a variable GOP structure would be useful, but it seems simply starting your GOP over at the scene change should be sufficient -- unless some players have problems with variable GOPs.

Now I'm feeling like I'll have to look into CCE because of its high praises. It just seems that (here) more people use TMPGEnc, so I assumed more people meant the best quality.

Lord Smurf said:
"I use 1-2-2 and 1-3-2 on live MPEG-2 captures from ATI.
I use 1-4-2 on other, which is max 15-length GOP.
1-5-2 is the max for a 18-length GOP."

Am I correct that the shorter GOP on the ATI captures is to reduce encoding time because you're MPEG encoding while real time capturing, but at all other times you use 1-4-2 which is slightly shorter than the commercial standard? This should give more I frames per second and improve transitions, but the quality of the I frames will be slightly less because there are more of them to fit in the same sized data stream?

So, would a person choose 1-4-2 if you're encoding at a higher bit rate (say 7000 Kbps or greater), but choose 1-5-2 if you're in the 5000-7000 range? Guidelines like that are what I'm looking for.

[Lord Smurf, as an aside, seeing you post here reminded me that every time I've tried to visit your web page using IE 6 SP1, I get a fatal browser error that crashes IE. No other web page has ever crashed IE for me. Do I have to upgrade to the absolute latest to view your page?]

Thanks.

[disturbed1]

Keep it simple, and follow the MPEG and DVD guidlines, which are 1/2 FPS with 2 B frames per P frame.

29.97 FPS / 2 = 15
23.976, 25 FPS / 2 = 12.

For low bitrates, multiple B frames, or an extended GOP length can increase the quality. Which is why VCD and SVCD both can have longer GOP lengths (per spec). SVCD tops out at 32, with no more than 2 B frames per P frame. I belive VCDs limit is 52 frames between I frames with no mention of B frame per P frame count.

I is an intra frame, recieves the least amount of compression, this is where the most information is stored.

P is a predicted frame. Recieves more compression than an I frame, but less compression than B frames.

B frames are bi-directional frames. These contain the least amount of information, and are needed for smooth playback.

I frames contain the whole picture information, B frames only contain the differences from the previous I frame, then the previous P frame. A P frame is predicted by the information conatained in the B frames.

B frames do the most work, and achieve the most compression.

Having too long of a GOP can cause seek and tracking errors in most mpeg decoding chips. Software players (your PC) usually won't have problems with this.

The reason behind a Max GOP length of 18 frames/36 fields is for easy navigation by standalone players, and better time tracking by the decoder.

Negating the B frames will cause less than stellar performance and quality unless a large bitrate is used (2MBytes/s DVD's limit is 9.8Mbits/s). Using all I frames uses more space. An I frame only MPEG file is almost the same as MJPEG.

There quite a few older GOP topics on this board, some really good information here.

Also google for "run length mpeg encoding" or "entropy mpeg encoding", MPEG.org has some great info and links. The older C-Cube data sheets were full of mpeg encoding facts.

[lordsmurf]

Lord Smurf said:
"I use 1-2-2 and 1-3-2 on live MPEG-2 captures from ATI.
I use 1-4-2 on other, which is max 15-length GOP.
1-5-2 is the max for a 18-length GOP."

Am I correct that the shorter GOP on the ATI captures is to reduce encoding time because you're MPEG encoding while real time capturing, but at all other times you use 1-4-2 which is slightly shorter than the commercial standard? This should give more I frames per second and improve transitions, but the quality of the I frames will be slightly less because there are more of them to fit in the same sized data stream?

So, would a person choose 1-4-2 if you're encoding at a higher bit rate (say 7000 Kbps or greater), but choose 1-5-2 if you're in the 5000-7000 range? Guidelines like that are what I'm looking for.

[Lord Smurf, as an aside, seeing you post here reminded me that every time I've tried to visit your web page using IE 6 SP1, I get a fatal browser error that crashes IE. No other web page has ever crashed IE for me. Do I have to upgrade to the absolute latest to view your page?]

Thanks.

On my site, the issue was in the code, and I'm altering it in the current update. The current index page and several others have no such code anymore. It should load fine with zero issues.

The farther you stray from the "I" the more messy it can get. With low bitrates, it was an issue of not giving the encoder too much data. In general, I'd say stick to 1-4-2 religiously. I only used 1-2-2 and 1-3-2 with ATI because of perceived performance and quality concerns.

I try to always use TMPGENC and PROCODER as my MPEG encoders, and both seem to do well at the scene change (TMPGENC needs option enabled).

Want to see bad GOP handling? Load WinDVR and capture with an MPEG-1 template. See the mess on scene change. Horrible.

[disturbed1]

I only used 1-2-2 and 1-3-2 with ATI because of perceived performance and quality concerns.

Just like to add.....

For Real Time Software encodes this will achieve better results than using the standard GOP lengths.

With the Real Time Software the necc. VLC DCT equations that are normally performed aren't really possible with real time software and todays CPUs. The B and P frames don't see enough CPU time and the quality is greatly degraded, using a longer GOP pattern will only give you more degraded frames.

When I still had my ATI cards I only captured AVI, or MPEG I frame 15Mb/s when space was a problem. But I migrated on to a more professional platform before MMC 7.1 (? I think) was available, so the quality could be better now compared to then.


To Lordsmurf -
No doubt you love your ATI's, as I did at one time too. I started with an 8mb Rage Fury Pro AIW 8) . Gotta tell you though, once you step up to a dedicated Real Time capture board you'll never look back. Canopus has a decent price on 2 products. Let's Edit RT+, captures DV from firewire and DV from anolouge input, with real time effects and real time output ($399), they also have the ACEDVIO with Vegas for $499, same card as the RT+, but Real Time effects are software only.

There nothing like having 2 layers of color correction with titles and 3d effects all displayed on you NTSC monitor in realtime while being played back from the timeline. Matrox has perhaps better cards for $100-$200 more but that's another topic

[lordsmurf]

Yeah, the ATI will serve it's purpose for my conversion years, but I want to move on here eventually to newer things. My old career choice wasn't working out, I got shot down just as I was really starting, so video may be a new avenue for me.

I've talked with lots of people, seen lots of demos and looked at lots of cards. The Canopus DV Storms and Matrox RT's all look nice. I'm leaning towards the older RT2500 because it's "only" $400 or so on eBay, but I'm not anywhere near that point in time where I can afford to buy one.

I'd love to hear your speech on whatever card you want to sell me on (send me a PM). I think the only Canopus aspect many of their cards look overpriced for what they do. Either that or Canopus sells themselves short.

Before ATI MMC 7.7 came out, I was actively pursuing other cards, as I wasn't getting anywhere with the older MMC software, and the cards just honestly don't work real well outside ATI MMC.

[BJ_M]

i'll add 2 cents worth....

first -- I saw someone mention about I frame insertion at scene changes and that main concept was listed in the list of encoders that dont do it ..

well - it does with the latest version IF you enable it in adv settings .. it even has two profiles for this .. it works very well.

canopus also does this -- but only on highest quality and mastering settings .. its doesnt do a great job on it on highest quality - but a prettty excellent job on mastering .. bow if only you could change the matrix table..

i go with lordsmurf on GOP -- pretty well what i would say .. i go with 1.5.2 for below 5000 and 1.4.2 over

for dvd and below bit rates -- IPPP not very efficient .. requires a higher bit rate for same quality as data bit rate is per GOP - not per frame .. therefore P frame use more bit rate than B frames .. there are also other types of frames besides IPB frames which rarely get mentioned (and are allowed in a dvd - but not used much as most software encoders dont produce them that i know of - but sony 's 35,000$ encoder does as do other hardware based encoders) ..

good for editing though -- IPPP is a standard GOP for sony MX and others for thier editing systems (some of them) ..

high end mpeg editing systems accually use any GOP .. because they use two buffers .. so it doesnt mater is IPP or 1.4.2 or 1.5.2


the first buffer decodes and feeds a compete video file to the second buffer .. the accual editing system only see's the output of the second buffer ..


B frames are forward and backward looking -- why 2 of them in a row doesnt mater ...

[disturbed1]

2 B frames together allow better encoding (quality) and smother decoding (playback). A decoder keeps the B frames in memory until the next P/I frame is seen. Giving the decoder 2 frames in memory.

Since B frames are bi directional, one B frame can refrence the previous GOP (non Closed bitstream of course) while the second B frame can reference the next P frame, giving the encoding a better interpretation of the motion vectors. Allows for a higher quality encode with a smoother playback.

Why not just use more B frames? It will overload the decoder. Since the decoder maintains all B frames in memory until a P frame or I frame is decoded, having too many B frames in a row will overflow the internal buffers.

If low delay is set in the sequence header, the decoder will keep no frames in memory. Since B frames are the only frames kept in memory, a stream without B frames is a low delay stream and not DVD Compliant.

[jal]

Hi,

I'm trying to learn ccesp 2.66 and get reasonable results going hi8->advc-100 dv-->mpeg2 for dvdr. So far the dv looks great on pc. When encoded doesn't look as good and on tv the hi8 cam to tv is much better than the dvdr. (washed out and somewhat blurry)

I was using default setting and cbr(ps). After reading this thread Im getting some degree of better understanding of IPB. The N and N/B you use to set the gop is confusing but the std 3 and 5 gives the ibbpbbpbbpbbpbb (1,4,2) I see recommended here. Also, from RTFM, I see a Qvalue <40 may give better results. For that I have to use vbr (pick vbr 3pass es vid). So how does 8500 max, 6500 avr, 2500 min with Q20 sound. I checked dvd compliant , since it will allow up to 9800 inspite of whats picked for max(right?)
Now when I get the resulting mpv and mpa where do I remux the files into a mpg . Using DVD WS2 for author, burn. Can I just do it right from there? If this works maybe cbr ps at 9700 will give quicker good results too?
Goal is to get a good quality recipe (where dvdr looks like s-vid cam into tv) and transfer a gazillion family hi8 and8s to dvds. efficiently.

(I posted something like this earlier today, but than found this thread)
Any suggestions and settings to ccesp appreciated. Also, will the good results be much easier to obtain when I start with a minidv cam? ie the loss Im seeing is just the nature of the analog to mpeg conversion.? Do people find video filters (at the dv stage) for sat and sharpness are standardly employed in this sit?

Thanks,
Jack

[W_Eagle]

I use 1-4-2 on other, which is max 15-length GOP.

Please assist my arithmethics. To my count, 1-4-2, which is IBBPBBPBBPBBP, is 13 frames. How come it's max. 15 GOP?

[SLK001]

One "I" frame,
four "P" frames,
and each one gets 2 "B" frames.

So, (4+1) x 2 "B" frames, plus the original "I" and "P" frames gives you 15.

Or,

IBBPBBPBBPBBPBB

[jal]

hi...to cont with my inquiry above a few posts....
retried same dv clip to mpeg ccesp with cbr9500, vbr 8500,7500,2500 (ah ws2 remuxes the files no sweat), and compared to previous vbr 7500. This time I rechecked that dvd qualities (somewhere) was high quality (which is was). Now since the max you can put it at is 8250, what will it leave alone and which streams will it reencode with the mpegnow codec??
The 3 where very similar but the cbr9500 seemed slightly better than the vbr 8500 (avg 7500) and both alittle better than the vbr 7500. Getting there, maybe.

a few other question about the cce settings. In ccesp, dvd compliant checked (a good thing?), the 'mpeg packets' = 2046.....seem to remember seeing ~4k with the mainconcept or tmpgenc. what's the function and significance?
Lastly, I don't know what to do if anything with the dither quant value and quantizer characteristics settings.

thanks,
jack

[BJ_M]

9500 is to high for dvd with audio -- at least it can be ....

9000 is better to use as max .. 8500 is even better .....

i dont care what dvd specs say is max allowed ...

[SatStorm]

@W_Eagle: With closed GOP

@All: Years before, Kwag and other enthusiasts of our hobby, use to test extremelly those things with the GOP structure, so did I.
After many tests, objervations, etc, I end up on what I used when I started: IBBPBBPBBPBBPBB (1-4-2) which is the full 15 GOP for PAL

IBBPBBPBBPBBP is the same but with closed GOPs. There are some authoring applications that work only with closed GOPs, but not TMPGenc Author, DVD Lab and almost all the new authoring applications.
Unfortunatelly, the Big Ones author their movies with some well known authoring applications, that only accept mpegs with closed GOPs sequencies, so that maybe explains why most commercial DVDs are IBBPBBPBBPBB. Actually, they are IBBPBBPBBPBBP, which is IBBPBBPBBPBBPBB closed.

IMHO, with Mainconcept and Procoder, you don't see any picture difference
when you use IBBPBBPBBPBB or IBBPBBPBBPBBPBB. You may see some difference with TMPGenc and probably CCE (which love short GOPs), but we talking for really low bitrates here: Less than 1850kb/s.
Why someone whoud encode 720 x 576 with such a low bitrate nowdays? And for 352 x 576, you have to go lower than 850kb/s to see quality change because of the used GOP matrix. But at this point, you have so low bitrate, that you have other problems to think about....
Those are my personal objervations of course, but after wasting a year testing Gops and Matrices, I end up to the default TMPGenc settings on this and honestly, I never had to deal with any kind of picture degrations that way! Overal, I believe that IBBPBBPBBPBBPBB gives good picture quality and fair compresion. With TMPGenc at least...

[SaSi]

Missed the thread for a few days.

SaSi, when you said most encoders (except CCE) "will almost religiously follow the IBBPBBPBB... structure and even try to encode scene cuts within B frames", are you saying if a scene change lands in a B frame that the encoder will attempt to do it by motion vectors (if that's the right words) rather than inserting an I frame at the scene change? If this is true, then I'm puzzled what good is the "Detect scene change" option in TMPGEnc (or any other encoder besides CCE)? I can see where a variable GOP structure would be useful, but it seems simply starting your GOP over at the scene change should be sufficient -- unless some players have problems with variable GOPs.

Your words exactly.

Actually, it's not that the encoders are doing a bad job. Mainly, the settings are a bit confusing for the users to set correctly. And no, I don't believe any decoder should have a problem with variable GOP settings.

In CCE, selecting "Equalize GOP length" (or something like that), will force the encoder to make all GOPs the same in structure and length. This means that inevitably, several scene changes will end up on a B frame.

In Tmpgenc, the setting Detect scene changes is there. I tried it and found that unless I use the Force Picture Type setting, i.e. scan the source file and have the program decide where to put I-frames, the encoder will still end up with scene changes on B-frames.

I viewed an encoded clip that had such scene changes fell on a B-frame frame by frame and realized that this is a major reason for "blockyness" and "pixelation". Once the new scene starts, it starts with macroblocks having visible boundaries (due to bitrate starvation). The next few frames show a decreased visibility of these macroblock boundaries. If no significant change follows, the blockiness blends in and disappears.

In high motion scenes (my benchmark source is Armageddon, the sequence where the shuttles dodge rocks and hit the meterorite), the blockiness is constantly apparent.

BJ_M, how do you enable the scene detection in advanced settings in Mainconcept? This is something I would like to try out.

[duhmez]

Dont forget you can adjust the gop based on the type of content. For example, when I encode Weekend update from Saturday night live, there is very little motion and ALOT of similar dat. In this case I lessen the p frames (more b frames then)
if it's fast action, upping the bitrate some and having perhaps one less b frame or something may be a good idea.

[BJ_M]

SaSi -- to enable scene change detection in main concept you will need the latest version (1.4.1 or better) -- as its not in prev. versions..


then in Details -> advanced -> GOP structure -> there is a drop down box with three choices (im doing this from memory as i dont have MC installed here) , pick the last one in the drop down (starts with a 'V' I believe) middle choice is a "fast" choice and top choice is non (default) ...