strict GOP structure for multiangle... anyone suggest an encoder?

[Mug Funky]

yo. i'm trying to multiangle something with a hardware encode i already have. only problem is it's GOP structure is:

IBBPBBPBBPBB closed

i've tried QuEnc, but it's "12-frame" closed gops look like:

IBBPBBPBBP, giving a 10-frame gop. TMPGenc does the same.

HC gives:

IBPBBPBBPBBP, which doesn't match either.

i'm aiming for 1 mbps - this is perfectly alright as it's just storyboards, so most of it can even be skipped blocks. quenc handles this content perfectly at CQ 3 and comes in at about 800kbps, but i can't use that because of how it closes GOPs :(

so... i'll mail a cookie to anyone who suggests an encoder that can get that GOP structure in a closed GOP, and do it reasonably well :)

[edit]

CCE doesn't do it either :(

[ronnylov]

I thought a closed GOP could not end with a B-frame. A closed GOP is a GOP that is not depending on next GOP to be able to decode it and then it can't end with a B-frame. What happens if you encode with open GOP instead of closed GOP?

[midnightsun]

^^ I think GOP's for multiangle must be closed as per DVD specs. Anyone correct me if I'm wrong. Anyway, your point on B-frames seems correct to me.

[hank315]

IBBPBBPBBPBB closedThis is weird, no encoder should do that.

Because you gave the GOPs for Quenc, TMPGenc and HC in temporal order I also assume IBBPBBPBBPBB is temporal.
Spatial: IPBBPBBPBB for Quenc and TMPGenc, IPBPBBPBBPBB for HC.
The spatial hardware encode: IPBBPBBPBBBB which is illegal for DVD.

Tweaking the temporal references and using one-way prediction for B-frames such a GOP can be done but IMO it's not the right way.

Do you have the temporal references of the frames of the hardware encode?

[Kika]

A closed GOP with B-Frames at the end? I have never seen something like that... :confused: How can this be? B-Frames do always refer to previous and/or following I- or P-Frames, so if you really have a closed GOP with a structure like that, the B-Frames must be... um... restricted. Weird ...

[Mug Funky]

Do you have the temporal references of the frames of the hardware encode?

umm... i'm not too sure what you're asking for there. i'm assuming the last two b-frames simply refer to the previous p-frame in the GOP they're in.

i agree it's not the best way to do it (even if the GOP ended in PPP it'd be a little more efficient). if you want i can take a look at the GOPs more finely in mpeg stream eye or something to see what else is going on.

btw, for anyone whose never seen GOPs like this before... anything encoded as closed-GOP on a spruce MPX3000 will be like this (and possibly anything encoded with the C-Cube chipset that the MPX3000 is based on). at the very least it means nearly all the region-4 anime out there, as we routinely encode closed-GOP.

it'd be great to have more control of the GOP structure on the Spruce card, but it doesn't give the option to tweak things that much - GOP structure is just a drop-down list.

it'd be totally awesome if i knew how to reverse-engineer the encoding application for this thing and plug in new ratecontrol... that'd really rock. of course i don't even know if the RC is done on the card or in software...

well, i think i'll work around it by capping the main feature and encoding it with a more sensible GOP structure. it's fiddly, but probably the best way.

[hank315]

if you want i can take a look at the GOPs more finely in mpeg stream eye or something to see what else is going on.mpeg stream eye would be fine, I'm just curious how the frames are ordered in the video stream.

And I don't think there are software encoders that will output such GOP's.

I don't know that much about merging multiangle clips but I always thought it would be OK if in both streams the GOP's start at the same frame and all GOP's are closed so the GOP structure itself isn't a real issue :confused:

[dragongodz]

just to hopefully clear some meanings up -

by spatial and temporal hank315 is refering to how the frames are stored in stream(spatial) and the order they are displayed in(temporal) which are not the same things.
i.e IPBBPBB in stream is actually displayed IBBPBBP.
this is of course because P frames must be done before the B frames or they couldnt be used as references.

in mpeg stream eye you have "decoder order"(temporal) and "stream order"(spatial) in the "Step by" section of the gui so you can see the 2 different modes in action. that is when you step forward a frame you can see which frame it jumps to depending which mode you have selected.

[Mug Funky]

ah, i get it about spatial and temporal. the GOPs i cited above are temporal (that's display order, right?). i got them just by stepping through virtualdubmod with the m2v loaded.

and after a lot of messing around with encodes, i conclude that (at least for authoring in DVDmaestro) you can't multiangle unless GOP structures are identical. it throws "Incompatible GOP Structure In Angle Block Detected" at me if i attempt to merge these clips.

ah well. i'm hacking a mac and a PC apart so i can capture an AVI off the digibeta (the mac would be a better option but the bloody thing is completely chocker-block at the moment, and i need about 200 gigs minimum). that way i can use the same settings for both angles, and take the Spruce's odd GOPs out of the equation.

however, i don't like my odds of getting it done tonight :( it's already 10:40pm on a sunday...

[dragongodz]

the GOPs i cited above are temporal (that's display order, right?)
yes. and that leads back to hank315's earlier post(post 4 of this thread). no i wont quote it all since people can easily just read it themselves. :)

i will say i dont know if its illegal or not. since maestro accepts it fine then it may not be. testing it on some other authoring programs(especially top end stuff like scenerist etc) would be interesting of course, also results from dvd verifier could be enlightening. i dont have them so unfortunatly can not.

even if not illegal though it does seem a strange thing to do since those last 2 B frames are going to be sub-optimal being backwards references only.

[bigotti5]

A closed GOP with B-Frames at the end?

http://forum.doom9.org/showthread.php?p=602367#post602367

[Mug Funky]

here's the first 2 gops of the spruce encode as stream eye sees them:
I B B P B B P B B P B B I B B P B B P B B P B B
stream order:
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
display order:
2 0 1 5 3 4 8 6 7 11 9 10 14 12 13 17 15 16 20 18 19 23 21 22

This gives:

BBIBBPBBPBBP BBIBBPBBPBBP, and it appears there's no references to outside GOPs. if i'm interpreting stream eye correctly (it's hard to read because on my machine it wants to put the video display on my TV), it appears the B-frames at the start of a GOP only contain backward references or intras, no forward refs. other b-frames contain forward, backward and forward-backward blocks.

@ dragongodz: if you'd like to check these streams for yourself just rip the feature off any of your madman discs - they're all done closed-GOP. it might be easier than my posting a sample :)

[hank315]

@Mug Funky
Thanks, the stream and display order you gave makes it clear how they did it.
Seems to be a normal open GOP stream and display order with the closed gop flag set and backward prediction-only of the first 2 B-frames (or intra), so only predicted from the I-frame.
If you want I can tweak HC a bit so it can produce such GOPs, will PM you.

[dragongodz]

ok i am back. just changed isp so offline for a while. :)

if you'd like to check these streams for yourself just rip the feature off any of your madman discs
will have to do that sometime. though not really needed now since you have shown what its doing.

[Trahald]

thats a weird gop stream , while technically closed its not something you see often. If hank makes hc able to do it i think it will be the only software that can. cce SP also makes traditional closed gops and does not encode closed gops with leading b frames. (not that it would be impossible for cce people to do it i guess, but the insentive isnt there as most in this predicament would just reencode the reference stream to a traditionally closed gop at a bitrate high enough to minimize loss)

[Mug Funky]

hehe... re-encoding m2vs is only something i'm prepared to do for OFLC discs (those buggers don't deserve quality).

anyway, i've grabbed the movie off the digi successfully (finally! everything that could go wrong did... and i'm still on my guard for blank frames that say "no video input" in last night's cap, as the SDI leads were crappy and it took 2 captures to get rid of the dropouts) and i'm going to do a re-encode with a sensible GOP structure tonight.

maybe i'll do an xvid of it too, just cause i get a kick out of encoding xvid off broadcast quality sources.

@ dragongodz: you're going to love this one, mate :) HDTV source, and that's a pure digital output from the animation software used. the only noise is from the original artwork.

[dragongodz]

dragongodz: you're going to love this one, mate
cool. ;)

EDIT:
hmm of course you havent told me what it is yet.

[Mug Funky]

oh yeah. well, it's some stupid cartoon about an old lady and some wizard with a robot house or something :)

in a completely unrelated matter, i'll be posting a problem sample soon that'll make all of your eyes spin. saggitaire might be interested (it's not uncorrelated noise, either).

[dragongodz]

well, it's some stupid cartoon about an old lady and some wizard with a robot house or something :)
HAHAHA thought it could be but didnt want to guess just incase. hmm of course she isnt really that old and the "house" is both mechanical and magical. glad to hear its going to be done from such a good quality source. now just to wait for it. :D

OT: for those wondering what the heck we are talking about Mug Funky encodes proffesionally. the "cartoon" he is talking about is a movie that i saw at the cinema, which he knows, and is going to be released on dvd here soon. once Mug finishes encoding and it gets authored and pressed etc etc etc that is. :D

[Rumbah]

OT: for those wondering what the heck we are talking about Mug Funky encodes proffesionally. the "cartoon" he is talking about is a movie that i saw at the cinema, which he knows, and is going to be released on dvd here soon. once Mug finishes encoding and it gets authored and pressed etc etc etc that is. :D
:eek:
I'm glad that this great movie is in such good hands. I've been waiting for the DVD release for months.

[dragongodz]

Rumbah - so you live in oz aswell and know exactly the movie we are talking about ? use its initials if you want to say its name. :)

[Rumbah]

No, I don't live in oz but in another PAL country. Here the initials for the movie are dws. But as google tells me the original title should be hnus and hmc in australia.

I may be wrong but that is the only film with an old woman, a wizard and a robot house that I know of. And I think that combination is not that common ;) .

[guada 2]

@Mug funky

IBBPBBPBBPBB/IBBPBBPBBPBB/IBBPBBPBBPBB

Have you tried CCE SP 2.67 ( Mpeg video setting option: M=3 & N/M=3) ?

[Trahald]

@Mug funky

IBBPBBPBBPBB/IBBPBBPBBPBB/IBBPBBPBBPBB

Have you tried CCE SP 2.67 ( Mpeg video setting option: M=3 & N/M=3) ?
actually i think you mean m=3 & n/m = 4.. but anyways.. he not only needs that gop structure but he needs it with closed gops. with closed gops you end up with ibpbbpbbpbbp ... (cce and hcenc anyways)


ibbpbbpbbpbb is only with open gops (normally). the gop confuses people sometimes since ibbpbbpbbpbb looks like the OPEN part is at the end but actually its at the beginning as the temporal order is bbibbpbbpbbp with the first b frame 'normally' referencing the prior gops last i or p frame. mugs gop is unusual as it uses this OPEN structure but is marked as CLOSED as the first b only backward references from the i in the current gop and does not use data from the prior gop.. it would lead to perhaps not as pretty of a B frame as usual but still decodable.

[guada 2]

Sorry Trahald,

Thank you for the information.

[dragongodz]

But as google tells me the original title should be hnus and hmc in australia.
correct. :)

of course if another company release it in your country it doesnt mean it will be Mugs encode. they may decide to do it inhouse etc.