QMatOp for DVD-RB : QMat Optimization [Archive] - Page 2

SAPSTAR

12th April 2005, 23:34

Originally posted by Axlemar
I wanted to try out the program, so I set it up like the instructions said and ran it for a while. Eventually I get the message:

execing cmd.exe /C move c:\documents and settings\joe\desktop\out\d2vavs\rebuilder.ecl c:\documents and settings\joe\desktop\out\d2vavs\rebuilder.bak
The syntx of the command is incorrect.

It seems to run fine for a while, but then this message just loops over and over again after about 30 minutes. Did I set something wrong? Using the newest version of both programs.

I know what it is !!! The spaces in the directory name...I never thought about it !!! Something I'll have to correct, before that you can try by changing your directory to a more "regular" one....without spaces I mean.

[EDIT] Should be corrected in the new current version 0.06b

SAPSTAR

12th April 2005, 23:39

Originally posted by Fishman0919
SAPSTAR, I've noticed something on a few movies done with QMatOP.
It looks like you have a 5% Tolerance but I have seen a couple final target perceptual error over that (6%-8%) and sometimes the final Iteration may be high in error (say 4.342245%) then the second to last one (say 3.524351245%)

just something I have noticed...Hope that helps

The 5% of tolerance is about the final size of the frame not the perceptual error. The perceptual error is not a percentage. Anyway, thank you for the info.

I may add some parameters in the .ini file to play with the tolerance percentage ?!? I was thinking about a parameter allowing to set the frame splitting rate (currently I take one frame every 20 seconds...) It could be changed....

Fishman0919

12th April 2005, 23:45

Very well.. THX for clearing that up....

Again THX for your work

Axlemar

13th April 2005, 01:15

This may sound like a stupid question, but do we have to do anything after it is done running and automatically closes? Q matrix is off by default so do we have to enable it or anything?

onesoul

13th April 2005, 01:56

@ SAPSTAR

QMatOp seems interesting but I have some questions.

Could you make sort of a guide explaining in more detail what is being done and what does the log describes as some other files created mean?

After the job done, for example, I have the intra and inter directories with some log files, why does appear 3 log files in each directory? What are the meaning of those values?
And I have the pics directory with 6 pics, 3 pics with "intra" description and other 3 with "inter" description, what does that mean, why only those 3 pics and why does the intra looks so much better?

And for last at anytime those new values for the matrix would fall off the standard ones?

I hope you don't mind all these questions, that's why a tiny guide would be appreciated.

Thanks

SAPSTAR

13th April 2005, 02:11

Originally posted by Axlemar
This may sound like a stupid question, but do we have to do anything after it is done running and automatically closes? Q matrix is off by default so do we have to enable it or anything?
Nope...The REBUILDER.ECL is modified to take into account the matrices.

SAPSTAR

13th April 2005, 02:16

Originally posted by onesoul
@ SAPSTAR

QMatOp seems interesting but I have some questions.

Could you make sort of a guide explaining in more detail what is being done and what does the log describes as some other files created mean?

After the job done, for example, I have the intra and inter directories with some log files, why does appear 3 log files in each directory? What are the meaning of those values?
And I have the pics directory with 6 pics, 3 pics with "intra" description and other 3 with "inter" description, what does that mean, why only those 3 pics and why does the intra looks so much better?

And for last at anytime those new values for the matrix would fall off the standard ones?

I hope you don't mind all these questions, that's why a tiny guide would be appreciated.

Thanks

I promise...as soon as I have the opportunity to do it...I'll do it.
Maybe this coming weekend...before I would be surprised.

First quick answer, the intra pics are better because they are less compressed than the inter....About the remaining logs and pics, in fact they could be deleted, they are the remaining of the last cell processed. The previous cells created the same kind files but were deleted after being processed. Anyway, I'll explain more, how it works in the promised document.

[EDIT] Just added a small pdf file in the last released zip file. (QMatOp006b.zip)

SAPSTAR

13th April 2005, 02:57

Added two new options in the .ini file to control the accuracy of the matrices. See my first message or my web site.(plus a small pdf file explaining a bit more)
[EDIT] RTF file not PDF
[EDIT] Don't play too much with the framesplitperiod parameter, as I just noticed it's not fully taken into account in the final result. So don't lose too much time by decreasing it until the next version is released.

sreg0r

13th April 2005, 05:08

i got an "There was an error opening this document. The file is damaged and could not be repaired" message when i tried to open the .pdf document.

SAPSTAR

13th April 2005, 05:12

Originally posted by sreg0r
i got an "There was an error opening this document. The file is damaged and could not be repaired" message when i tried to open the .pdf document.
Thank you for the info..The pdf file must be corrupted...I'll create another one later. Sorry.

[EDIT] I recreated the ZIP file and put a rtf file instead of a pdf file.

onesoul

13th April 2005, 14:11

Hi, I read the rtf file but I still need to ask if the inter matrix is the non-intra matrix? (I don't really know what are the differences between intra and non-intra matrix :))
-Normalization of the previous result. (Matrices are moved if under the MPEG2 standard) Do you mean matrices are changed or recalculated accordingly the MPEG2 standard?

Cheers

SAPSTAR

13th April 2005, 14:21

Originally posted by onesoul
Hi, I read the rtf file but I still need to ask if the inter matrix is the non-intra matrix? (I don't really know what are the differences between intra and non-intra matrix :))
Do you mean matrices are changed or recalculated accordingly the MPEG2 standard?

Cheers

Inter matrices = non-intra matrices, they are mostly used in the P and B frames.
Matrices are moved according to the MPEG2 standard, because in the MPEG2 standard the first "cell" in the intra matrix must be 8, and in the inter must be at least 8. (it means the base level in MPEG2 is 8)

Example : My first cell has a 6 value, I will move by 2 the whole matrix, in order to be standard. Some people may say why don't you multiply instead of moving the matrix. I say, you're wrong guys by multiplying you will amplify the steps between the details tresholds.

SAPSTAR

14th April 2005, 04:53

IMPORTANT CORRECTION IN THIS RELEASE NOBODY SHOULD USE THE 0.07b VERSION !!! :scared:
An error was introduced in the 0.07b, the average is wrongly calculated. :(

[EDIT] Now I'm freezing the current version for few days, I want to test more my versions before releasing them from now, too many stupid errors were introduced in the past.

Axlemar

14th April 2005, 05:51

Its cool. Thanks for the quick update.

Fishman0919

14th April 2005, 10:41

@SAPSTAR, THX again.

PS Damm u, :D just finished run QMatOp 0.07b over night w/ 1%t and 1sf to wake up to a warning not to use ver 0.07b....lol that will learn me not to use the max settings...joke :D

SAPSTAR

14th April 2005, 11:36

Originally posted by Fishman0919
@SAPSTAR, THX again.

PS Damm u, :D just finished run QMatOp 0.07b over night w/ 1%t and 1sf to wake up to a warning not to use ver 0.07b....lol that will learn me not to use the max settings...joke :D
WOW !!! Must have been long ? With 0.08b, I ran a 1% tolerance with 10 seconds period...and I found it long....let me know the results if you ever try again the "1 1" with 0.08b. Sorry again for the error !:scared:
[EDIT] By the way, I just noticed a memory allocation part where I think you could have some pb with a framesplitperiod of 1. If I were you, I would stay above 4 or 5 until I correct this possible memory allocation error.

ernstblaauw

14th April 2005, 14:40

Why is jdobbs not participating in this discussion? Does he not like this program?

SAPSTAR

14th April 2005, 16:11

Originally posted by ernstblaauw
Why is jdobbs not participating in this discussion? Does he not like this program?
Don't know...Ask him :D

Fishman0919

14th April 2005, 16:19

rerun 0.08b w/ 1%e and 1st and got an error... couldn't catch what it was but something like:

Error: java/...lan.......245........
Error: java/...lan.......246........
Error: java/...lan.......247........

over and over...but it did propuce a matrix

hope this helps

SAPSTAR

14th April 2005, 16:51

Originally posted by Fishman0919
rerun 0.08b w/ 1%e and 1st and got an error... couldn't catch what it was but something like:

Error: java/...lan.......245........
Error: java/...lan.......246........
Error: java/...lan.......247........

over and over...but it did propuce a matrix

hope this helps
See the EDIT of my previous reply to you...I noticed a possible memory overflow in the case of a very low FrameSplitPeriod, I think that's what you just encountered, until I correct it, you should stay above 4 or 5.(depending on the biggest cell you have)
I think that the result when overflowed just take into account the first 100 seconds of the current cell.

And thank you very much for you testing, it helps, because it confirmed what I feared about this possible memory overflow.

Fishman0919

14th April 2005, 16:54

Ya, I noticed your post and wanted to let you know...

Re-ran 0.8b with 5% + 5st and no errors, process took about 2 hours on an AMD 64 3800+

SAPSTAR

14th April 2005, 16:59

Originally posted by Fishman0919
Ya, I noticed your post and wanted to let you know...

Re-ran 0.8b with 5% + 5st and no errors, process took about 2 hours on an AMD 64 3800+
Good ! Again thank you for your testing :thanks:...FYI I'm now sure that it's only related to the framesplitperiod, you can keep the tolerance at 1, it will work.
Lucky man, you have a fast computer !!!! Your encoding must be very fast too !

Fishman0919

14th April 2005, 17:03

I'll do Family Guy Season 3 Disc 3 with 1% and 3st.. see now that goes

Lucky man, you have a fast computer !!!! Your encoding must be very fast too !

about an hour for 2 passes with CCE.

SAPSTAR

14th April 2005, 18:25

Originally posted by Fishman0919
I'll do Family Guy Season 3 Disc 3 with 1% and 3st.. see now that goes about an hour for 2 passes with CCE.
Let me know the quality of the result, plus did you choose any specific matrix in DVD-Rebuilder settings ? (If yes, they influence the result).
I'm thinking about including an option to indicate if it's a "NORMAL" movie or a cartoon type.....just to help the internal algorithm, currently a way to "help" it, is to choose a "smart" matrix in the DVD-Rebuilder settings.
Example : if you chose a CG matrix, the optimization will be for a CG kind of movie. If none was specified, it just does its best by itself.

Fishman0919

14th April 2005, 18:56

I'm encoding the movie now, play some of the .m2v files and they look real good.

no I didn't choose any specific matrix in DVD-Rebuilder settings ..just encoder default

If you like I can run Family Guy on a diff machine (got 4 to play with) and choose "Smooth CG" in DVD-RB let you know the diff.

Edit: I got alittle time on my hands right now, went sking two weekends ago and got blindsided by a 10 year old girl who was going out of control... I was standing still wait to go over a hill and she hit me from behind...broke my left leg in 2 spots. Anyway lucky for laptops and wireless connections...hehe

Axlemar

15th April 2005, 04:31

If you don't mind, I am going to ask questions along with my results:

I backed up the same animation dvd twice using the newest cce sp 2.70, once using rebuilder 84.1 by itself and again using qmatop .08b with 2% tolerance and 10 seconds. Both times I used default matrix settings in rebuilder. When I compared the two outputs on several different players, qmatop had introduced a considerable amount of static on its output that was originally 5.47 gigs in size (low compression rate). I also noticed that the audio skipped on some of my players during high action scenes from the qmatop version while the rebuilder version didn't (probably the result of the player not liking QMat). To try and make sure it wasn't a bad burn, I backed up another disc from the same animation series that had the same previews, and the previews skipped in the same places and on the same players as the first one did. So overall: Some players played it fine and the final output size was good, just more static was introduced than with the default rebuilder matrix settings with QMat disabled. I will run another test with the same source using 1% tolerance and 5 seconds.

Now my questions: Is this supposed to replace CCE version of QMat by formulating and adjusting matrices through the series of programs you have set up? Also, could my settings have introduced the static because they were too strict or not strict enough. I have copied animated series using rebuilder with CCE QMat enabled and the output didn't have the kind of static I get from qmatop. Thanks.

(Also about using different matrices in rebuilder, I heard using things like the cg matrix uses cce filters that can cause very strange results. The best quality would come from having QMatOP analyze every second with a low tolerance so that the original picture's available bitrate is optimised across the detail in the image right?

SAPSTAR

15th April 2005, 11:37

Originally posted by Axlemar
If you don't mind, I am going to ask questions along with my results:

I backed up the same animation dvd twice using the newest cce sp 2.70, once using rebuilder 84.1 by itself and again using qmatop .08b with 2% tolerance and 10 seconds. Both times I used default matrix settings in rebuilder. When I compared the two outputs on several different players, qmatop had introduced a considerable amount of static on its output that was originally 5.47 gigs in size (low compression rate). I also noticed that the audio skipped on some of my players during high action scenes from the qmatop version while the rebuilder version didn't (probably the result of the player not liking QMat). To try and make sure it wasn't a bad burn, I backed up another disc from the same animation series that had the same previews, and the previews skipped in the same places and on the same players as the first one did. So overall: Some players played it fine and the final output size was good, just more static was introduced than with the default rebuilder matrix settings with QMat disabled. I will run another test with the same source using 1% tolerance and 5 seconds.

Hmm interesting, what I would like to know first, did you do any test with the first version ??? I ask that because, in the first version, quantifiers rarely went above 190, and I heard some players may have difficulties with very high Quantifiers, I may have to chnage my "normalization" function in order to come back closer to usual MPEG2 standard ?!? For example, I'm not sure if all players are able to handle an inter matrix non starting by a value between 8 and 16.....Even if the pure MPEG2 standard doesn't specify anything about that.

About the static, what do you exactly mean ?

Now my questions: Is this supposed to replace CCE version of QMat by formulating and adjusting matrices through the series of programs you have set up? Also, could my settings have introduced the static because they were too strict or not strict enough. I have copied animated series using rebuilder with CCE QMat enabled and the output didn't have the kind of static I get from qmatop. Thanks.

QMatOp replaces the matrices define in each cell, in the ECL file. If you chose any matrix before, it will be taken into account.(less if you increase the framesplitperiod). I don't think your settings are the problem. I'll check with the framesplitperiod parameter, I may have introduced a bad side effect, I may have to tweak a little more this part.

(Also about using different matrices in rebuilder, I heard using things like the cg matrix uses cce filters that can cause very strange results. The best quality would come from having QMatOP analyze every second with a low tolerance so that the original picture's available bitrate is optimised across the detail in the image right?
Yep exactly...for the intra part, but I'm still looking for a way to improve my inter (non-intra) part. This part is rather difficult to optimize. I have an idea, but I fear the processing will be too long.

Axlemar

15th April 2005, 18:45

Sorry, I never used the first version so I can't say if it would fix the skipping. I am in the process of doing everything with 1% tolerance and five seconds just to see if there is a difference. The values in every matrix started at 8 and then varied, but some went up to as high as 242 (I need to learn more about this kinda thing. I don't know if that is significant or not). On some of my players, QMat done by CCE SP without QMatOP made the images kinda jumpy (once again the compat with QMat), but some played them fine and there wasn't the kinda static that I got, so this process must be doing things much different then CCE. By static, I mean that the overall image, especially during movement, had a lot of small breaks in it that kinda looks like static. If some objects moved, you can notice the space around them kinda ripple and break. Usually I see similar results in very low bitrate compression with CCE, but the source in my test was only about half way in 5 gig.

As for improvements, if you think that you may have found a way to optimize your program in anyway that will yield higher quality results, even at the cost of time, I encourage you to do them. The point of including another program is to spend extra time improving the output anyway, and besides you can always include options to change the mode if people don't want to wait. By the way, what exactly would the -m switch do in dctune? Thanks for your hard work.

(Ok I finished a 1% tolerance and 5 seconds. The quality is a bit better, the motion breaks (static) aren't as appearent. It still skips in the same places on the same players (which I guess it should cause the matrices are calculated in the same fashion with the same ranges. When you fix the memory problem, I will run a 1% and 1 second to see if there is a big difference or not. Thanks).

SAPSTAR

16th April 2005, 18:06

I just released the last version of QMatOp, I corrected some bugs (see README.txt), now you can run with a 1s framesplitperiod.
I also added the possibility to automatically start the encoding after the processing. (you have to keep dvd-rb open...)
I totally changed the way inter matrices are processed, it will be a little bit longer, but I expect a lot of quality from this one.

A special greeting to FishMan0919, who helped me a lot by testing it deeply....thank you man !

Fishman0919

17th April 2005, 10:36

Very nice SAPSTAR.... like the autostart of DVD-RB

Carpo

17th April 2005, 14:02

so what do ppl suggest for

tolerance=x
framesplitperiod=x

atm i have it as 1 and 5 - is this too much/too little ?

SAPSTAR

17th April 2005, 14:59

Originally posted by Carpo
so what do ppl suggest for

tolerance=x
framesplitperiod=x

atm i have it as 1 and 5 - is this too much/too little ?
If you have a fast computer, I would say :
tolerance=1
framesplitperiod=1

I have an AMD 2600, so I keep framesplitperiod=5 for my current tests.
tolerance can be 1, as the runtime doesn't change so much with it...

Axlemar

17th April 2005, 22:12

I ran the newest version at 1% tolerance and 5 seconds and it still produces too many ripples and breakes during motion for me to use it over the quality of rebuilder and CCE SP alone. Can someone else please back up an animated series with and without this program and compare them?

Fishman0919

17th April 2005, 22:21

ME...ME...oh ...I did... I did a few of the the Family Guy Season disc with and without QMatOp testing for SAPSTAR and the new ver look just a little better (to me on my 65' HDTV) then without... I did a few DragonBall Z and GT disc with QMatOp and got the same results.

onesoul

17th April 2005, 22:36

@SAPSTAR

I don't want to sound depreciative towards your work, I find it an interesting tool. I tried 1% and 4s split on monsters_inc. I encoded the resulted matrices, aswell the mpeg standard, on the referred cell, it was very disappointing, the calculated matrices made the encode with too much artifacts that didn't show up on the other one.

I will try sometime 1% and 1s, but I feel this tool isn't yet fully optimized, maybe because of dctune?

Cheers

Axlemar

17th April 2005, 23:49

I am having a similar problem with "artifacts" appearing more in the final encodes. I have tried 1%/5 and 1%/1 and it still seems to have more artifacting then without using it, which is odd because dctune sounds like a pretty cool piece of technology. Maybe it has something to do with the conversion from avs to avi and maybe something is being mapped incorrectly?

Fishman0919

18th April 2005, 00:14

onesoul and Axlemar, what encoder are you using???

I found with HC and QuEnc I got some MacroBlocking in some movies but with CCE SP 2.70 and w/adaptive Q-matrix switching turned on there was some noticeable improvement in quality then without using QMatOp. I didn't try any encodings w/ CCE SP 2.67 but it does not have adaptive Q-matrix switching so results may be the same as with HC and QuEnc.

SAPSTAR

18th April 2005, 01:30

@onesoul
Of course, it not yet fully tuned...it's released for about one week :)
Anyway, it doesn't really come from dctune it self, dctune is only optimizing the matrix for a frame given several parameters. I'm still trying to figure out the best settings, especially for the inter matrices. The intra part is rather easy, I think that it's already good. The pb is for inter matrices, they are used ofr the encryption in P and B pics...which are difficult to assess, that's why in the last version I tried to work with delta frames (differences between succesive frames...)...But I'm still trying to find the best parameters for them.
@axlemar
avs2avi is only used to interpret some AVS scripts which are creating the frames, no conversion is really performed.....I didn't want to rewrite an AVS interpreter.....

@all
When giving some issues, could add more info :
Kind of movie (animation , regular), PAL/NTSC, encoder, matrices specified or not specified in DVD-RB (remember that's a way to help QMatOp, myself I always select some smooth matrices in DVD-RB before running QMatOp for animations, it gives a clue to the processing).

I really hope obtaining some really good final results, I will have to improve it but I need help for that. Especially for the inter/non-intra matrices, this part is really driving me crazy....yesterday I tried several possibilities without success :(

Up to now, I have the feeling that the pbs are mainly with animations, correct me if I'm wrong....I need to know.

Axlemar

18th April 2005, 04:17

I will have to try it with some other sources before I can tell you if the problem is with animation or not, but animation is almost the only thing I back up. I always use CCE SP 2.70 (the newest verison with the newest patch) and almost always ntsc.

I was told earlier that you don't have to enable the QMat switch and that QMatOP handles it for you (I guess from inserting the matrices directly into the ecl file?). I never change the default matrix cause I don't like the idea of using filters on the source and I know it is possible to achieve good quality without them. It would seem that the biggest problem is between succesive frames (the motion breaks I tried to describe) like you said you were working on.

What would happen if I entered the QMat switch in rebuilder? Would it do QMat without using QMatOP's values or would it just use QMat's values but sway them depending on the scene or something?

I wish I could help you on a more technical level, but I have limited knowledge on how I can apart from testing sources.

SAPSTAR

18th April 2005, 11:27

Originally posted by Axlemar
...I was told earlier that you don't have to enable the QMat switch and that QMatOP handles it for you (I guess from inserting the matrices directly into the ecl file?). I never change the default matrix cause I don't like the idea of using filters on the source and I know it is possible to achieve good quality without them....
BEWARE !!! You always use some QMat in any MPEG2 encoding !!! That's the baseline of the MPEG1/2/4 algorithm. By default, CCE uses its own matrix, which is, from my point of view not really good for any kind of movie/animation.(mosquito noise) The usage of a specific QMat is NOT the usage of a filter!!!!! The QMat is used to tell what details it has to encode and what details it can ignore, as the MPEG2 codec is NOT a lossless encoding method , it's based on the fact that the human eye can't really some levels of details.(levels of difference of light in fact)
So my purpose is just to try to find the best matrix for each encoding instead of using a default average matrix.
Even if you don't use QMatOP, I encourage you to use at least some animation dedicated matrices in the future.....

Fishman0919

18th April 2005, 11:52

Axlemar, QMatOp and adaptive Q-matrix switching in CCE SP are two diff thing (kinda of the same but diff).... QMatOp analyzes each cell that DVD-RB creates in the "Prepare" phase and then QMatOp creates an optimized matrix for that cell from a set of parameters (avr bitrate, max bitrate....). Adaptive Q-matrix switching in CCE SP analyzes each GOP (Group of Pictures) it is encoding and adjusts the matrix it is using so to give maximize video quality. Having adaptive Q-matrix switching turned on in DVD-RB w/CCE SP will give you better video quality but some (very, very few) DVD Players do not like it...

from the CCE SP guide...

Disable adaptive Q-matrix switching
Cinema Craft Encoder SP (later than 2.67) adjusts quantization matrix
automatically to improve the quality of image when Multipass
VBR is selected. However, some DVD player does not support such
MPEG streams as encoded with adaptive quantization matrix, and
may cause the block noise when playback. When you encounter the
problem, you can avoid it with this option.

What is adaptive Q-matrix switching ?
There are some cases that Cinema Craft Encoder SP cannot keep
the minimum bitrate. It is considered to be caused by that the lowest
quantization scale cannot raise the bitrate under a single quantization
matrix. To solve the problem, Cinema Craft Encoder SP automatically
creates other matrices cutting the value of each element in the
matrices by half, quarter and so on, and applies another one to produce
a much lower distortion for much higher bitrate. This switching
of quantization matrix can occur at every GOP.
Thus, adaptive Q-matrix switching can improve the quality of image,
especially with lower bitrate sources or CG works or the edge of
fading out.

If you have been using CCE SP 2.70 with DVD-RB for awhile you have had adaptive Q-matrix switching on because it is turn on by default in CCE SP, just resently jdobbs turned it off by default in DVD-RB with an option to turn in on.

Hope this helps

onesoul

18th April 2005, 12:04

@ Fishman0919

Doesn't make much sense activating the adaptive Q-matrix option when using QMatOp because a complete new set of matrices will or could be used. So QMatOp would do no much difference I think.

I used HC, on pal anime Monsters_Inc.

edit: @ SAPSTAR

I tried an earlier version 0.05 I think, but didnt actually encode but I noticed that the generated inter pictures looked better than on this later version. It already not looked so good with artifacts but on the later it is blackened, almost can't see it.

Fishman0919

18th April 2005, 12:23

@onesoul

Yes a new matrix will be created but base on what matrix is being using... so if you have a matrix closely optimized for that cell... maybe not perfect yet but close... CCE SP will bring it to what it needs to be when it needs to.

SAPSTAR

18th April 2005, 16:17

Originally posted by onesoul
@ SAPSTAR

I tried an earlier version 0.05 I think, but didnt actually encode but I noticed that the generated inter pictures looked better than on this later version. It already not looked so good with artifacts but on the later it is blackened, almost can't see it.
If you are talking about the generated pics in the QMatOp tree, that's perfectly normal !!! As inter pics are no more real frame but delta frames (difference between successive frames) as inter matrices are used to encode P/B pics !!!!
So you can't deduct anything from what you can see in the QMatOp directory.
[EDIT]I just finished calculations (on an old sheet of paper :) ), and found a new formula, which may improve things especially concerning the non-intra part, and I really hope it will make the artifacts disappear. I forgot to take into account the GOP structure, in my previous versions.