Log in

View Full Version : Max overflow improvment and degradation question


Zep
30th December 2003, 11:33
I see in beta 3 the default values for both
Max overflow improvment and degradation are 5%.

Wasn't the default values in beta 1 and 2 60%?

that is a huge difference and I'm curious as to how
this affects encodes. Even more so how this affects
encoding short 1 minute clips and the ability to hit
target sizes.

Can someone please explain this in detail to me? :)


thanks

Mr_G_123
30th December 2003, 11:46
i have enocded in diffrent matrices (hvs best picture,mpeg,heavy compression) and when it is set to 5% it almost doesnt hit the target size, in my case target: 5000 kb result: 6000 kb.
basicly in koepi explanid options 26/9/2003 that was posted here the overflow is explaind and in www.xvid-gr.tk (i think it is down for a while).

Zep
30th December 2003, 15:19
Originally posted by Mr_G_123
i have enocded in diffrent matrices (hvs best picture,mpeg,heavy compression) and when it is set to 5% it almost doesnt hit the target size, in my case target: 5000 kb result: 6000 kb.
basicly in koepi explanid options 26/9/2003 that was posted here the overflow is explaind and in www.xvid-gr.tk (i think it is down for a while).

yes I'm seeing that too. my 1 minute clips are 10 megs to hit
but endup 2 megs under or over. I still do not know why
sometimes over and others under.

Looking at debug view Xvid is not choosing the correct Q for the frames. when under sized i see Xvid choosing Q's that are too high
instead of using the bits it has and could use. I assume if i bump
the 5% then Xvid could adjust faster.

thanks for the link. Does it work for you? I get 404 error
at that link. (currently)


thanks

Koepi
30th December 2003, 15:44
Well, read doom9's xvid-1.0 guide andsee the explanations there.

The defaults are well chosen for full length encdings and not clips. you want to set it to 10%/10%/10% (min/max overflow degradation/improvement and overflow control strength).

This will lead to a wider varying quantizer range, but this is obviously what you want/need for these small samples.

Regards
Koepi

Mr_G_123
31st December 2003, 09:55
i have done some tests (i will publish them soon) with diffrent matrices (hvs , bulletproof's) and also the most important the min/max overflow and control strength. my conclusion was that the best combination is hvs-best-picture/Bulletproof's Heavy Compression
AND: min/max overflow both 30% , overflow control strength : 100% (i know it is very extreme but in some cases it works like 0%)

you will see better when i post the status windows of the test soon (many status windows :)).

and koepi is right see the doom9 xvid1.0 guide (i have seen him and it explains wonderfully:D )
http://www.doom9.org/xvid-vdub-final.htm

Zep
31st December 2003, 19:54
Originally posted by Koepi
Well, read doom9's xvid-1.0 guide andsee the explanations there.

The defaults are well chosen for full length encdings and not clips. you want to set it to 10%/10%/10% (min/max overflow degradation/improvement and overflow control strength).

This will lead to a wider varying quantizer range, but this is obviously what you want/need for these small samples.

Regards
Koepi

Thanks I will give that a try and check out the guide.

Zep
31st December 2003, 19:55
Originally posted by Mr_G_123
i have done some tests (i will publish them soon) with diffrent matrices (hvs , bulletproof's) and also the most important the min/max overflow and control strength. my conclusion was that the best combination is hvs-best-picture/Bulletproof's Heavy Compression
AND: min/max overflow both 30% , overflow control strength : 100% (i know it is very extreme but in some cases it works like 0%)

you will see better when i post the status windows of the test soon (many status windows :)).




please do that sounds very interesting :)

Mr_G_123
4th January 2004, 11:22
http://207.68.162.250/cgi-bin/getmsg/ultimate-30-30-100.jpg?curmbox=F000000001&a=c02db49705f40bd9fac43439f79683a1&msg=MSG1073211034.0&start=290452&len=540352&mimepart=3&disk=207.68.162.69_d329&login=wozgreg&domain=hotmail%2ecom&hm___sig=5e328f4601c3d2ab9b2238b838cca3983a02ad30aac949de

Mr_G_123
4th January 2004, 11:30
i am very sorry, the edit post kinda of screwd my post with puting images from sites
http://win.mail.ru/cgi-bin/readmsg/ultimate-30-30-100.jpg
please tell me (the stupid me) how to post pictures from my hard drive and not from sites because my photos are in my mail. please help me:confused:

Koepi
4th January 2004, 11:51
Uplod the images to soe webspace and link to there.

You should remove that link above, that leads to a passport site - i doubt you want the people to look into your account (if that's possible with passport).

regards
Koepi

Mr_G_123
5th January 2004, 08:41
http://www.asdfhost.com/members/mrg123/
that is my site, and by the end of this week i will upload the test to there.
thank you thank you thank you thank you thank you thank you thank you
koepi for the help.:D ;) :D ;)

Mr_G_123
6th January 2004, 11:45
i have problems with posting the test(by the way, he was shrinked due to the fact that it takes very very very long time to make and document the test) so i will give the final result:
the best setting for 2 pass are:
min overflow and max overflow both 30%, and control stregth 100%
Why?
to give full control to xvid (100%) and a medium touch to the frames: below 30 % and the target size not achived, more than 50% and xvid will kill the movie

Zep
6th January 2004, 13:19
Originally posted by Mr_G_123
i have problems with posting the test(by the way, he was shrinked due to the fact that it takes very very very long time to make and document the test) so i will give the final result:
the best setting for 2 pass are:
min overflow and max overflow both 30%, and control stregth 100%
Why?
to give full control to xvid (100%) and a medium touch to the frames: below 30 % and the target size not achived, more than 50% and xvid will kill the movie

wow that seems high for 1 hour + encode

did you try 10 10 10 as Koepi suggested?

I will have to do some more tests for sure now.

acrespo
6th January 2004, 17:17
I set to 5/5/10 and target bitrate 1080 KB/s to give me a 175 Mbytes file size. The result was a 220 Mbytes file size. I will try other settings.

acrespo
7th January 2004, 14:46
Originally posted by acrespo
I set to 5/5/10 and target bitrate 1080 KB/s to give me a 175 Mbytes file size. The result was a 220 Mbytes file size. I will try other settings.

I didn't know what settings was wrong in my system, but after load defaults and change some settings (cartoon mode,trellis etc), the file size is correct.

I think the matrix was wrong or any other setting. I don't know.

sapient
10th January 2004, 06:05
Has something changed in the way the overflow parameters work in beta3? In what way? Why were the default values changed so drastically? Is doom9's explanation in his guide still valid? I tried searching the forum for answers but couldn't find anything. In any case, what are the recommended values for full movies?

Zep
10th January 2004, 14:25
Originally posted by Koepi
Well, read doom9's xvid-1.0 guide andsee the explanations there.

The defaults are well chosen for full length encdings and not clips. you want to set it to 10%/10%/10% (min/max overflow degradation/improvement and overflow control strength).

This will lead to a wider varying quantizer range, but this is obviously what you want/need for these small samples.

Regards
Koepi

I tried 10%/10%/10% on a 44 minute HDTV encode with target
size of 350 megs and it was 3 megs under.


beta3 is just not as good at hitting sizes as beta1
(not sure about beta2 i didn't use it much) beta1 i was
always around 500K under only.

Mr_G_123
11th January 2004, 14:37
use 30%/30%/100%, it is great in hitting target size with a very good result