Andreas GOP 1-3-3-1
 

Hello everybody, I am following now for a while all the discussions on this forum and I am very impressed with all the results that have been achieved.

One thing I would like to bring to all of your attention, is something that Andreas is bringing up in his recent article about the KVCD-template in the German forum. Andreas mentions that with the GOP 1-18(12)-3-1 structure he notices some "pumping" movement in the encoded file.

Well, I experience the same and have followed his advice to use a GOP structure of 1-3-3-1 instead of the current KVCD 1-18(12)-3-1.

By doing this, the "pumping" movement has disappeared in my encodings against a little file size increase. Andreas has done some testing himself and he gives some examples of some file size comparisons:

1. Standard (352x576 mit 320x320 ma) aus dem KVCD Forum (1/15/3/1) und eine min Datenrate von 300kbps, max. 2000kbps Video, CQ68, Ergebnis ist 97,6MB

2. Alles gleich bis auf GOP mit 1/3/3/1 und Matrix, Ergebnis ist 98,3 MB

I only do Divx-encodings and do not know if that is the reason for GOP 1-18(12)-3-1 to cause the "pumping", but the alternative GOP works for me. I have been able to do Planet of the Apes with the modified GOP and still fit the whole movie into one CD-R 80 with the KVCD 352*240 plus template.

Hope this is useful to some of you that experience the same problem

Hi Turbo:

Thanks for the comments. Andreas article is based on the evaluation of the old GOP that had the last number, the MAX number of frames per GOP, set to 0. This meant that the GOP could be extremely large, because TMPEG would make the decision as to how long it could make the GOP.
The latest KVCD templates have been changed to integrate our own Quantization Matrix, and a GOP with a MAX size of 48 ( 50 for PAL ).
The "pumping effect" he refered to, was that every time a new GOP was inserted, the newly fresh I frame caused a "flash" or a brightness increase in the video. Specially in low lit and long static scenes. It was almost not noticeable in action scenes. The "pumping" effect is not present anymore in any of the latest KVCD templates, because they all use the same GOP. With the new KVCD Q. Matrix and the new GOP, it's a completely different ball game :wink:

-kwag

Hi Kwag,

The encodings I recently did where I had the "pumping" effect were done with the latest KVCD-template having a GOP with a max size of 48. By using your current template with the only modification being the GOP structure 1-3-3-1, I get a GOP with a max size of 16 which my player (a Samsung HTDTL-70) seems to like more. The "pumping" effect I mean, is not as what you describe a "flash" or a brightness increase. It's more like a non-fluent movement which is gone by using the modified GOP 1-3-3-1.

Anyway, it works for me and the file size increase is only about 1 to 2%.

Thanks a lot for your outstanding templates and keep up the good work!

Aye, its very visible in still scenes with low contrast backgrounds(dim lit walls aso.)

Aditionally to the the 1-3-3-1 GOP I use an raised minimum bitrate of 700 Andreas propagated. Its another 5% increase in filesize but its well worth it, both in quality and compatibility(some DVD players cant handle bitstreams lower than ~700 Kbit)

Btw: Regarding compatibility. Ive changed kwags 704x488(576) template to 480 horizontal resolution and mpeg2. With the other changes mentioned above its far more compatible(in fact it should be in the limits of the SVCD standard) and the option for all the fancy SVCD stuff like subtitles and an second audio stream. Eg. I got 6th Day on 2 CD-Rs with both english(160 kbit) and german(128 kbit) sound channels.

cheers

Rookie

Hi Turbo:

Is this effect on a PAL systems?
Because I can't reproduce this effect on any of my encodes with the NTSC tamplates.

Thanks,
-kwag

It should not, at least the same effect shows on an test encode with your NTSC template I just made. As far as I remember the mpeg specs the problem is the image degeneration due to the long GOP. The P frames arent encoded lossless so the longer the GOP the more blurriness and blocks will show. The longer the GOP the more obvious is the insertion of an new I frame.
As I said the effect is only seeable in low contrast scenes with an next to uniform background.
The picture is sort of jumping all 2 seconds(for PAL)

Hi Kwag,

I use your NTSC-templates and play them in my Samsung HDTL-70 DVD-player which plays both NTSC and PAL. As I live in Europe, my television is based on the PAL-system, so yes, it could have something to do with PAL. By the way, I do no conversion to PAL in the encoding.

The thing is, with the GOP 1-5823-3-1, I have the non-fluent movement I already talked about and I do not have it with the GOP 1-3-3-1. With the last one, I have a very smooth screen movement.

It can also be seen with PowerDVD but less than on my television screen.

Rookie,

Do you use a min bitrate of 700 just because of the compatability or do you really see a quality increase? I would think that by using the CQ-encoding mode this has no effect on the quality, but it will, just like you said, increase the filesize.

For CQ mode the quality increase is obvious in still scenes. With 300 Kbit the image is sort of fading and wobbling(mostly not very visible on my TV set) but for CQ_VBR its not cause CQ_VBR next to never goes down to the minimum bitrate(from the limitied experience I have with that mode so far) Anyway I do up it to 700 and adjust the quality slider to be on the save side.

good night

Rookie

Hello guys,
Please do NOT use PowerDVD to preview. The mpeg decoder in PowerDVD is about the worse I've seen. I only use WinDVD for test purposes, and what you see in PowerDVD, the "trailing streaks" and colored blocks that appear to "grow" in the image, don't show at all in WinDVd or in any standalone DVD player I've tested.
I just encoded a couple of 5 second clips with our GOP of 1-5823-3-1-48 and another one as 1-3-3-1-48, and with the 1-3-3-1-48 the file size is almost the same, but there are more "mosquito" effects around borders than with the 1-5823-3-1-48. The longer GOP "smoothes" the sharp edges, and they blend in better with the picture. I started the first frame exactly at the point where Andreas took the screenshot in the link far below. On a longer sampled encode, the 1-5823-3-1-48 will show a much smaller file size than in very small samples.
Here are the video streams for your analysis.
http://ns1.shidima.com/kwag/test-1-3-3-1.m1v
http://ns1.shidima.com/kwag/test-1-5823-3-1.m1v
Both are 704x480 with the KVCD's Quantization Matrix.
Please compare the above samples with the screenshot here:
http://www.dvd-svcd-forum.de/phorum/...t=ST;f=18;t=46

Obviously, the screen shot above, was not encoded with our Q. Matrix :wink:

-kwag

BTW: Here's the same sample at the lower resolution of 352x240, also with the PLUS template.
There are "barely" visible artifacts.
http://ns1.shidima.com/kwag/test-352x240-1-5823-3-1.m1v

-kwag

Hi Turbo,

Did you see these effects with the new templates at the default settings?
Because I see these exact effects, but only if I decrease the CQ value.

Hi Kwag,

I really appreciate that you took some time to do some testing yourself with the GOP 1-3-3-1 and I downloaded your samples.

In my standalone DVD-player on my television-screen, I can not see any "mosquito" effects at all with both GOP's in my subjective observation. However, it is not the "mosquito" effect or "trailing streaks" and colored blocks I am refering to. It is really only the the movement of the images, (it looks like some frames are skipped, or "jump" into another) although both of your samples look fine this time so perhaps it has something to do also with the quality of the source, which in my case is Divx.

Please do not understand me wrong, I am not trying to say that GOP 1-3-3-1 gives a better general result than the current one. I am only saying, that it works better for me in my situation.

Anyway, I will be out for a couple of days but when I am back I will do some encodings. If you are interested, I'll let you know the results or make some samples available.

Muaddib,

I use the standard template with the default settings. By using GOP 1-3-3-1, the "pumping" movement disappears in my case.

:twisted: DivX :twisted: You should have mentioned that from the beginning :lol:
That's the problem right there :wink:

-kwag

Kwag,

I have mentioned from the beginning that it was DIVX[/quote]

My first Post!!!!!!!!!! :D
I have a Panasonic DVD-RV26. I'm having the same problem with my movies with the movie sort of jumping or like the frames slow down or popping. 8O But it's only happens every so often. It happens at the begining of Hart's War (ntsc) when the train is moving. It's like the still pictures stop for a second then start again,but the sound works good and the video keeps playing. I can watch the movie fine but like I said the popping or what ever it is happens only a couple of times throw the movie. I used both new templetes 352x480 regular and the plus, and had the popping in both templetes. And I watched the movie on 27" tv. Meathed to create movie: Smartripper,Dvd2avi,Fitcd,Tmpg,Nero. Any help on fixing this popping or what ever would get great. :D

Hello kwag,

I read this diskussion advertent and with a big surprise.

Well, and I thing kwag see the same, we have no “war” with the different systems setting and between PAL and NTSC and the result of our tests (I hope). We and many other peoples looking for the best results for the defined problem “one cd for a whole movie”.
O.K. in europe we have e.g. PAL. I can only test for this settings and I take the Ideas from kwag to my tests. All comparitive tests are working with the current templates from the KVCD-Forum at the time I testing. In the second part of the KVCD article (22 Juli 2002) I used the templates that I download from the 21 Juli 2002 of the KVCD-Forum. You could see the working templates on this examples:

1. KVCD-Forum PAL

http://www.digitalfaq.com/archives/error.gif
Picture 1, original KVCD template (352x576) 21 Juli 2002




2. KVCD-Forum PAL „plus“ bzw. beta

http://www.digitalfaq.com/archives/error.gif
Picture 2, original KVCD plus template (352x576), 21 Juli 2002




3. KVCD Beitrag Teil1 CQ68

http://www.digitalfaq.com/archives/error.gif
Picture 3, original „article part one“ template (352x576, 320x320ma), 02 Juli 2002

http://www.digitalfaq.com/archives/error.gif
Picture 4, original „article part one“ template (352x576, 320x320ma), 02 Juli 2002




One of the big problems is to find a sequence that is significant to show us the disadvantages of our settings. Most we change only one setting in the whole template and then we compare it with the reference setting we found further. I tested it on three short sequences from three differend films. To test it on a bright area mostly you can´t see realy a difference. Bright sequences are good to for tests to see how sharp is the current resolution with slow to very slow motions. Some block artifacts if you have quick movings (e.g. a fireball) and light Moskitoartifacts around edges.

On darker sequences like the examples above you see how “dirty” are the walls and consistent surfaces. You can see it better on the night because the pictures are dark. If you have more color and small structures on surfaces the encoder have to do a hard work. Picture 2 and 4 show only a moving of the two persons. The walls left and right “standing still”. But in the original KVCDplus telmpate 21 Juli (it´s the better template) you have many moving blocks and colorartifacs. One of a compressing problem is that any kind of colors are changing from the original to a color it looks like purple if you don´t refresh the original block. The I-frame is the picture that includes all exactly informations for every pixel and finaly for every macroblock. If you have only one I-frame every 2 seconds and the other pictures are only P and B frames you lose more and more correkt basic information through the bachward conversion of the iDCT.
You can see it if you download the examples. On “my” (not realy my it´s for everyone) PAL template you have “moving areas” too but with less coloratifacts and less “moving surfaces”. And if you change the minimum datarate to 650kbps then the compression advantages of a big GOP is over.



I thing the basic differences are the resulutions from PAL and NTSC and the frames rates. In PAL we have around 20% more of datas in one secound of a movie. Another reason why I used a small GOP is to move close to the DVD convertions of a GOP bigness of 15. On the TV screen you see less mistakes than on a PC monitor. Both together close to DVD and lower mistakes on the TV are the main reason that I try to used a small GOP.

ANDREAS

[/quote]
Yes you did :oops: Sorry!
I did't see it :roll:

-kwag

Kwag,
I encoded hart's war (ntsc) again to test somethings. What I tested was the beginning when the train is moving throw the forest. I see the picture like pause for a second or so or like the still picture is showed for a second, and then the movie plays on like nothing happened. :cry: I see it when the movie starts and if you look in the top left corner were the forest is and then the movie moves onto the train. It only happens about a couple times throw the movie. And yes it happens on other dvd rips. I tried vbv=0 but the movie play bad :cry: tried, vbv=50 didn't see any difference :cry: ,tried vbv=48 and 600 low bit rate didn't see any difference :cry: . So should I try the gop of 1-3-3-1 or do you have any other ideas. 8O The templetes are great :D and the picture is great :D ,but I have the picture like pause for a second or two about a couple times throw out the movie. Could it be because I'm running the dvd player throw the vcr? :?: or am I encoding something wrong? :?: I didn't change anything in your templete except to 1:1 vga,and I have the new templete (downloaded 7/21/02). If you need any more info, I post in this post on page 1. Thanks for any help.

acidfire,
One or two times in a complete movie 8O
Have you tried another media?
Because if you said that it happened every so many seconds, or a couple of times per minute, then I would try other parameters in the template. But not for two times in a complete movie!

-kwag

Kwag,
Thanks for the reply. :D I will try to do a couple more movies and see what I come up with. I might just be too picky. :roll: But I'm sure it's not your templete! :D One more question. Do you recommend useing temperal smooth (2,2) for dvd's? I have read some say yes and other say no. Thanks for all your help and great work on your templetes. 8)