RB v1.26 bitrate redistribution

[Boulder]

You could first do the redistribution with the standard matrix and then use RB-Opt to change it to whatever you want. The redistribution doesn't depend much on the matrix as can be seen in some tests in the RB-Opt thread.

[Voodoochild]

Hi
I've done my first Movie with the redistributed feature.
1. I'm very pleased with the results, Many Many Thanks for the new feature.
2. I've noticed in some segments that the bitrate drops below 2000, example 2,923 Kbs ----> 1,825 Kbs. I looked at the result and it was dark scene, so it came out good, but on the edge...

My question is, wouldn't it be wise in those few cases to take some bitrate from high scene let's say above 5000 and give a bit to those scenes below 2000 manually through the DVD-RB segment viewer, just to be on the safe side, that the scene won't be blocky ?

10x
Elad

[blutach]

Perhaps use the segment editor for that Voodoo or do a quick encode using RME?

Regards

[Sharc]

@Voodoochild
Just for my curiosity: Which encoder / matrix did you use?

[Voodoochild]

@Voodoochild
Just for my curiosity: Which encoder / matrix did you use?

I used the default cce Matrix for all segments with 25% sample
cce sp2.
I will try now with avama6, using the technique of adding a bit more to those below 2000, to see if three any difference in that scene. I will also check the scene I took some bitrate, to see weather the quality declined, though I don't think it will, as long as it stays above 80% compression.

Perhaps use the segment editor for that Voodoo or do a quick encode using RME?

Regards

Currently the last RME doesn't work.. I did sent Rockas the problem in the mail.

10x Elad.

[Voodoochild]

After testing again same DVD, with added bitrate in scene that were below 2000 br to 2500 and taking from scene that were above 80% compression and from end credits (no point having end credits with br above 2000) . I didn't see any differences at all while looking at scenes that demanded high br and the one that I added br to 2500 from 1800.
Still just to be on the safe side I prefer to add bitrate to those scenes below 2000.

10x
Elad

[jdobbs]

MHO: It's a waste to add bits to a section the doesn't need it. The redistribution gives you what it takes to keep the quality current. So if you have one section that has 5000Kbs in use and another that has 1800 -- it is telling you that you are going to get the same quality out of the 1800 (on that segment -- because of lack of demand) that 5000 would give you (in the other -- because of high demand).

The result of manually changing it: The 1800 would be at a higher quality level than the rest of the stream, and the 5000 would be lower.

Demand means a lot. In the example after the manual change, the segment that was changes from 5000 to (say) 4800 would acually have lower quality than the one that changed from 1800 to say 2000

[Sharc]

I second that. I did similar tests with scenes that got bitrates as low as approx. 1100 kbps.
As you have been using the default matrix there was little risk that CCE didn't do the redistribution properly - that has actually been my concern.

[Voodoochild]

......
Demand means a lot. In the example after the manual change, the segment that was changes from 5000 to (say) 4800 would acually have lower quality than the one that changed from 1800 to say 2000

I see your point jdobbs and logically it's true, but I've noticed in allot of my backups that when a movie is above 80% compression I can't tell the difference from the original at all.
So I thought if I have a scene that has 5000 and the segment viewer tells me it's 86%, for me it means that I can take 4% off for example and still won't tell the difference from the source. By giving those 4% to a very low bitrate scene that has below 2000 I can maybe avoid the tiny chance that this scene will suffer luck of BR. I only did it to 1 scene and couldn't tell the difference from the original 1800 BR. So in my case what you're saying is true.


10x
Elad

[Carpo]

to those that are getting low Q values what version of avisynth are you using ? in a few tests i did using 2.5.6a and 2.5.7 - i found that low Q values were given out more under 2.5.7 that 2.5.6a.

One film under 2.5.6a gave Q38 - and under 2.5.7 Q24 was given, this was repeated using a number of films (all PAL) and using the encoder default matrix

[Boulder]

Would you mind uploading a small sample of the same frames that have been encoded using Avisynth 2.5.6a and 2.5.7 using the exact same Q? Some people have reported sizing issues with 2.5.7 but no one has provided samples to show what really is happening there.

[jdobbs]

to those that are getting low Q values what version of avisynth are you using ? in a few tests i did using 2.5.6a and 2.5.7 - i found that low Q values were given out more under 2.5.7 that 2.5.6a.

One film under 2.5.6a gave Q38 - and under 2.5.7 Q24 was given, this was repeated using a number of films (all PAL) and using the encoder default matrix
This is the Q used for redistribution? If so, I'm not sure how that is possible -- as it is determined by the size of the source and the total frames. AVISYNTH isn't a player in the determination of the Base_Q.

[Sharc]

@carpo:
Are you on CCE SP or CCE SP Trial?
Did you in DVD-RB select both "Enable bitrate redistribution" (Mode menu) and "One pass VBR (w/analysis)" (Settings menu) at the same time? This would involve Avisynth for producing a sample file for OPV - still, it wouldn't explain why you get different values for Q.
If you want to do redistribution, only the "Enable bitrate redistribution" should be selected, the "One pass VBR" should not be selected.

[Carpo]

This is the Q used for redistribution? If so, I'm not sure how that is possible -- as it is determined by the size of the source and the total frames. AVISYNTH isn't a player in the determination of the Base_Q.

yes that is correct - same film same - same settings - diff avs version - due to my isp monitoring upload stream i doubt i can upload - although i know this would help they have a wired AUP - they seem to think if you upload more than a few hundred meg you must be a heavy user and cap you, maybe this is an error close to home as i do need to re-install soon - maybe then things might sort themselfs out

[Carpo]

@carpo:
Are you on CCE SP or CCE SP Trial?
Did you in DVD-RB select both "Enable bitrate redistribution" (Mode menu) and "One pass VBR (w/analysis)" (Settings menu) at the same time? This would involve Avisynth for producing a sample file for OPV - still, it wouldn't explain why you get different values for Q.
If you want to do redistribution, only the "Enable bitrate redistribution" should be selected, the "One pass VBR" should not be selected.


see above - same settings same dvd source, same everything apart from avisynth version, well apart from defraged drive ;)


cce sp 2.70.02 - and droping back to avisynth 2.56.a has also yeied a speed boost

[Options]
redistribute=1

is the only thing that is changed

have done with and without over two diff versions of avisynth

[Boulder]

although i know this would help they have a wired AUP - they seem to think if you upload more than a few hundred meg you must be a heavy user and cap you, maybe this is an error close to home as i do need to re-install soon - maybe then things might sort themselfs outYou don't need to UL hundreds of megabytes, even 10-15 frames might do which would mean a couple of megabytes - if there is a difference between the two Avisynth versions, it should be observable in pretty much every frame.

[Carpo]

has anyone had cce freeze when doing the redist pass ? i have had it happen on 2 segmants on a test disc of mine - cpu hits 50% on one core mem is about 256meg, i notice is does this when the gop changes from open to close, if i kill cce on the segmant, it pops back up doing the next one in the list

[Sharc]

I never had CCE freezing. I had CCE silently aborting with high bitrate matrices (like FHE) and low Q's - apparently a CCE bug as discussed previously in this thread. I did however never see CCE aborting with the Q calculated by RB-Opt ("final Q"). Perhaps just by coincidence, and I was just lucky.

[Carpo]

this isnt aborting it sits there at 66% on the two segmants and thats it, mem stays steady at 260meg and its using 50% cpu time, have reripped the disc, ran it thru fixvts, trying dvdshrink on it now, and will give the drives a stern checking just incase, but seeing as i have done 5 other dvds without issue im not overly convinced that its a drive issue

[Sharc]

......The redistribution doesn't depend much on the matrix as can be seen in some tests in the RB-Opt thread.
... only - at least for CCE - as long as Q is selected for equal target sizes (means "final_Q" for DVD-5). However, this "final_Q" depends largely on the selected matrix, for example Q=75 for FHE, Q=34 for CCE default, Q=18 for Avamat6, to name a few. If this applies only for CCE and if this is related to the CCE bug - I don't know. Anyway, this "final_Q" is obtained by RB-Opt and yields distribution profiles which are in fact almost independent from the particular matrix. Just my experience from many tests.

[Voodoochild]

Hi All.
I'm trying to back up A Sopranos DVD, which has 4 episodes.
Compression level is quite High 58%. I thought this is classic for redistribution, so I did prepare phase cce sp2, 20% sample, default matrix, q=43. When looking at the results, I noticed that I have 4 peeks; each one is the 1.35 minuets of the beginning trailer. Wouldn't it be wiser since it's only the beginning trailer, not the episode it self to lower those high value back and give more to the rest of the segments that are actually the episode it self?

attach a graph of the redistributed vs orginal


10x Elad

[Boulder]

A change request to the HC redistribution : use the GOP options that are selected for the encode process. Now it seems to use 12-2 but I'd much rather use autoGOP with HC.

[tom942]

I've been doing several tests with HC and I would like to know what others do with LUMGAIN.

If I use LUMGAIN=4 in prepare and encode phase, I can see little squares or blocks in dark scenes and also in the clear ones.

Then I set both phases LUMGAIN=0, and the image improves a lot in both cases. In the clear ones, I can´t notice any blocks but in the dark ones, I can still see blocks (less than before).

Would it be a good idea to do the prepare phase with LUMGAIN=0 and later, the encode phase, with LUMGAIN with 1-4?

I ask this because I've seen that depending if you've got LUMGAIN OFF or ON, the bitrate distribution changes, and then I don´t if I could be doing a wrong thing :confused:

[jdobbs]

I personally set LUMGAIN to 2 -- I think setting it to 4 is probably too much for most situations. If you do "2" it does a good job of preventing blocking in dark transition scenes.

[laserfan]

I'm trying to back up A Sopranos DVD, which has 4 episodes...Wouldn't it be wiser since it's only the beginning trailer, not the episode it self to lower those high value back and give more to the rest of the segments that are actually the episode it self?I made a lot of Sopranos backups prior to the RB Redistribution feature. The opening sequence looks like absolute crap in the 4 episode discs, but in all cases the main show looks just fine.

Probably the BEST solution is to edit-out the openings altogether, and give all the bits to the rest of the episode(s). That opening challenges the process beyond anything in the actual shows themselves.

If you like the openings it's prolly best to just leave BR off.

[blutach]

Hi voodoo,

If you're talking about the bit with the "Woke up this morning" song at the start which is the same in all eps, I have a different idea if you wanna keep it.

Given all eps are in the same VTS, change the VCID of the segment in eps 2, 3 and 4 to match the VCID in ep 1. The excised VCIDs will become unreferenced material, which can be eliminated with VobBlanker (I expect DVDRB would also eliminate it).

The only thing that you would notice is a lack of seamlessness in eps 2, 3 and 4 (the layer break flag should be on) - big deal if you can save 5 minutes of challenging video.

If you wanna pursue this, the easiest way is to do it in PgcEdit. Would take no more than a minute.

Regards

[tom942]

@jdobbs

Thank you. I'll give it a try.

Do you use LUMGAIN=2 always, or depends of the source?. And do you use the "encoder default" matrix or use other matrices that compress less?.

Could it be other idea to set LUMGAIN=0 and set CPU=4 in MPEG2SOURCE_OPTS?

[Voodoochild]

Hi voodoo,

If you're talking about the bit with the "Woke up this morning" song at the start which is the same in all eps, I have a different idea if you wanna keep it.

Given all eps are in the same VTS, change the VCID of the segment in eps 2, 3 and 4 to match the VCID in ep 1. The excised VCIDs will become unreferenced material, which can be eliminated with VobBlanker (I expect DVDRB would also eliminate it).

The only thing that you would notice is a lack of seamlessness in eps 2, 3 and 4 (the layer break flag should be on) - big deal if you can save 5 minutes of challenging video.

If you wanna pursue this, the easiest way is to do it in PgcEdit. Would take no more than a minute.

Regards

Excellent idea blutach, I will give it a try,
P.S, in which program do you mean by saying "the layer break flag should be on".. ?

10x Allot
Elad

[jdobbs]

@jdobbs

Thank you. I'll give it a try.

Do you use LUMGAIN=2 always, or depends of the source?. And do you use the "encoder default" matrix or use other matrices that compress less?.

Could it be other idea to set LUMGAIN=0 and set CPU=4 in MPEG2SOURCE_OPTS?I use the default matrices on almost everything. I change the matrix only when I have a specific reason for doing so.

[blutach]

Voodoo - see your PMs. The cell flag will need to register the non-seamless nature of the cell and the discontinuity in the system time clock for eps 2, 3 and 4 (flag value = 2). PgcEdit should do this for you.

Regards

[laserfan]

If you're talking about the bit with the "Woke up this morning" song at the start which is the same in all eps, I have a different idea if you wanna keep it.That IS an interesting idea blutach, but note the lead-in is NOT the same--the titles change to reflect actor appearances and (at least) writing and directing credits for each episode as well. At least one episode I recall too that the "lead-in" did not start at the beginning.

I do find the opening credits interesting, particularly when I see for example that Michael Imperioli has written the episode we're about to watch. We also like the DVDs' "last week on...", the written synopsis of the episode we're about to see, and then afterward the "next time on...". I know others that just pound thru the episodes without looking at ANY of that stuff...! To each his own, there is no accounting for taste, YMMV, etc. etc. The great thing is that we have all these tools to do what we want!

[Voodoochild]

... great thing is that we have all these tools to do what we want! ...


Amen for that :)

[blutach]

For mine, if you can save 5 minutes of video, and the tradeoff is some bloke's name in the writing credits, I'd save the video (particularly if, as voodoo as said, compression is high and presumably quant is too). But each to their own as you say.

Regards

[laserfan]

OTOH I even like that opening HBO logo with the dramatic music and the noisy video screen--but boy does that wreak havoc on my Sigma-based media players (can't show it without stuttering)!!!!

Now I know how to manipulate or edit-out these things, but for me "what's done is done" ie. at some point you have to decide when to stop messing with it and just sit and enjoy it... :)

[Carpo]

are people getting their dvds on target when using redist? of the five i have done 2 were 4.32gb 2 were 4.34gb and 1 is 4.30gb - is this normal or it depend on the film

[jdobbs]

Actully mine are pretty much coming in at 4.33 -- which is very, very close to the specified size. I have seen a couple that are smaller, but usually when the reduction is fairly small.

[Insomniak4700]

Has anybody so far tried using CCEAQM=1 in the rebuilder.ini file and use redistribution at the same time? (I set the redist percent at 100)

My eyes tell me the results are quite good, but I'd like to know if anybody actually did any tests, with data?

[Fishman0919]

I use CCEAQM=1 with all my encodings with CCE SP/SP2.
No problems at all.

[archaeo]

I use CCEAQM=1 with all my encodings with CCE SP/SP2.
No problems at all.



Same here, w/no problems. Sizing has been on target, too.

[jdobbs]

I've been using it (on-and-off) also with no problems.

[Carpo]

hmm i was always told not to use that setting as some players dont like it - although i have done one or two and the cheap (at the time) £50 player i bought parents plays them ok and the pc does - should i reconsider this option ? and wouldnt this option effect the redist ? or is it totally safe to use CCEAQM=1 and redist ?

[Fishman0919]

as some players dont like it

Some 1st gen DVD Player (Sony DVPS-3000/7000 some problems and Pioneer DV-09 same if will play DVD+/-R... Pioneer DV-05 seems fine but has issues playing DVD+/-R's sometimes) and a few 2nd Gen Players.

[jdobbs]

Custom Technologies warns in their documentation that it may be incompatible with some players -- that's why I add that warning when we talk about it... but there are a lot of people using it, and not many who have run into problems. I thnk Fishman0919 might be right -- it's probably something that only affects old players.

[Carpo]

one i bought my parents is old and not a top brand make and it seems to plays the ones i did with the ccq setting on - but to the other Q how safe is it with redist option, suppose i could do a few dvds and see :)

[jdobbs]

It works fine with redistribution.

[tom942]

With Bitrate Distribution enabled, has it got sense to set the bias as I ask in this thread?

http://forum.doom9.org/showthread.php?t=128884

[Insomniak4700]

Thanks for your opinions everyone, I did a couple of discs with both CCEAQM and redist, both my players seem to play them back fine, and sizing was spot on.

[jamewoong]

Did anyone notice a huge difference when enable the bitrate redistribution?

Between, high or low Base_Q is better?

Thanks.

EDIT:
This feature is really ugly... Very disappointed with the result.

Without redistribution:
552 mins - quality ok.

With redistribution:
786 mins - quality is okay, but sometime, blocky image come up. Also, when the image is dark, it will make it worst. When the image is clear (light), the quality is better.

Using Procoder

[Carpo]

have you tried doing the same encode but using HC or CCE and seeing if it still looks bad ? it could be procoder having a fit were as other encoders may be ok

[jdobbs]

Did anyone notice a huge difference when enable the bitrate redistribution?

Between, high or low Base_Q is better?

Thanks.

EDIT:
This feature is really ugly... Very disappointed with the result.

Without redistribution:
552 mins - quality ok.

With redistribution:
786 mins - quality is okay, but sometime, blocky image come up. Also, when the image is dark, it will make it worst. When the image is clear (light), the quality is better.

Using ProcoderWas the blocky output on a very small segment (like less than 500 frames)? I noticed that on one I did where there was a fade-in to an announcement. It was caused by the fact that only the blackness (before the fade-in) was used as the sample. I've made some adjustments for the next release to prevent that.

As I've said before, this feature is something that needs to be tested and proven. While I haven't seen negative results (beyond what I just mentioned), I have seen positive on several discs.

...786 mins...That's a lot of extra time for doing a sampling??? You didn't change anything else?