Quantcast Avisynth: Death of Stmedianfilter - Page 4 - digitalFAQ.com Forums [Archives]
  #61  
05-21-2003, 09:16 AM
jorel jorel is offline
Invalid Email / Banned / Spammer
 
Join Date: Aug 2002
Location: Brasil - MG - third stone from the sun
Posts: 5,570
Thanks: 0
Thanked 0 Times in 0 Posts
hi friends.

i'm testing the new script with little variations.(seems great)
a big, big tests,lots of samples.

please,wait for my results.
thanks

Reply With Quote
Someday, 12:01 PM
admin's Avatar
Site Staff / Ad Manager
 
Join Date: Dec 2002
Posts: 42
Thanks: ∞
Thanked 42 Times in 42 Posts
  #62  
05-21-2003, 02:50 PM
Wolfi Wolfi is offline
Free Member
 
Join Date: Oct 2002
Location: Sweden
Posts: 489
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via AIM to Wolfi Send a message via MSN to Wolfi Send a message via Yahoo to Wolfi
If I'm getting a smaller CQ value with the newest script, what am I doing wrong then

//Wolfi
Reply With Quote
  #63  
05-21-2003, 02:51 PM
Kane Kane is offline
Free Member
 
Join Date: Jan 2003
Posts: 398
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by Wolfi
If I'm getting a smaller CQ value with the newest script, what am I doing wrong then

//Wolfi
what was your "older" script
how much was the cq decreased?
__________________
greetz Kane
Reply With Quote
  #64  
05-23-2003, 06:13 AM
Wolfi Wolfi is offline
Free Member
 
Join Date: Oct 2002
Location: Sweden
Posts: 489
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via AIM to Wolfi Send a message via MSN to Wolfi Send a message via Yahoo to Wolfi
Oldest script was

ConvertToYUY2()
LegalClip()
GripCrop( width=544, height=480 )
GripSize(resizer="BicubicResize")
STMedianFilter(10, 30, 0, 0, 10, 30)
SpaceDust()
unfilter(50,50)
temporalsmoother(1,2)
mergechroma(blur(1.50))
mergeluma(blur(0.2))
DctFilter(1,1,1,1,1,.5,.5,0)
GripBorders()
Letterbox(0,0,32,32)
LegalClip()

Newest

ConvertToYUY2()
LegalClip()
unfilter(50,50)
GripCrop( width=544, height=480 )
GripSize(resizer="BicubicResize")
STMedianFilter(8, 32, 0, 0, 8, 32)
SpaceDust()
temporalsmoother(1,2)
mergechroma(blur(1.50))
mergeluma(blur(0.2))
DctFilter(1,1,1,1,1,.5,.5,0)
GripBorders()
Letterbox(0,0,32,32)
LegalClip()

I does'nt drop more then 1-2 but I thought cq would rise a bit

//Wolfi
Reply With Quote
  #65  
05-23-2003, 11:41 AM
Kane Kane is offline
Free Member
 
Join Date: Jan 2003
Posts: 398
Thanks: 0
Thanked 0 Times in 0 Posts
maybe it´s because of the changes in STMFilter
with the new settings , the cq is not that higher, as with the old settings
__________________
greetz Kane
Reply With Quote
  #66  
05-23-2003, 05:38 PM
Wolfi Wolfi is offline
Free Member
 
Join Date: Oct 2002
Location: Sweden
Posts: 489
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via AIM to Wolfi Send a message via MSN to Wolfi Send a message via Yahoo to Wolfi
But the picture gets sharper and clearer with the new script, right?

//Wolfi
Reply With Quote
  #67  
05-23-2003, 05:41 PM
Kane Kane is offline
Free Member
 
Join Date: Jan 2003
Posts: 398
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by Wolfi
But the picture gets sharper and clearer with the new script, right?

//Wolfi
yes, it does (but i still don´t use STM )
just make a short sample with both scripts and compare
__________________
greetz Kane
Reply With Quote
  #68  
05-23-2003, 05:49 PM
Wolfi Wolfi is offline
Free Member
 
Join Date: Oct 2002
Location: Sweden
Posts: 489
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via AIM to Wolfi Send a message via MSN to Wolfi Send a message via Yahoo to Wolfi
Thanks Kane but I'm not so good to compare clips Should I look at walls, faces, background details or what Please point me in the right direction...

//Wolfi
Reply With Quote
  #69  
05-23-2003, 05:55 PM
Kane Kane is offline
Free Member
 
Join Date: Jan 2003
Posts: 398
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by Wolfi
faces, background details

//Wolfi
yes, that is what i look at, when comparing script

are the background and faces less blury, can i see more details?
how do action scenes and dark scenes look?
something like that

if a script is really better than the older one, you will notice the difference, just by viewing , let´s say 10sec of each script one behind the other
__________________
greetz Kane
Reply With Quote
  #70  
05-23-2003, 05:58 PM
Wolfi Wolfi is offline
Free Member
 
Join Date: Oct 2002
Location: Sweden
Posts: 489
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via AIM to Wolfi Send a message via MSN to Wolfi Send a message via Yahoo to Wolfi
Are you comparing with "full screen" or just normal? Thank you Kane

//Wolfi
Reply With Quote
  #71  
05-23-2003, 06:04 PM
Kane Kane is offline
Free Member
 
Join Date: Jan 2003
Posts: 398
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by Wolfi
Are you comparing with "full screen" or just normal? Thank you Kane

//Wolfi

normaly i burn a view samples on a cd-rw and view it on tv, ´cos this is, where i want to watch it
sometimes i use windvd fullscreen
__________________
greetz Kane
Reply With Quote
  #72  
05-23-2003, 09:02 PM
jorel jorel is offline
Invalid Email / Banned / Spammer
 
Join Date: Aug 2002
Location: Brasil - MG - third stone from the sun
Posts: 5,570
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by jorel
hi friends.

i'm testing the new script with little variations.(seems great)
a big, big tests,lots of samples.

please,wait for my results.
thanks

yeah, i quote my own post for the same reason...
the results of my sample tests.

was hard to do and to write,you know my "klinglish"
then...

sizes and times using the
full new script and variations:

George Harrison sung in "any road":

"traveling there and traveling here
everywhere in every gear...
and i've been traveling throught the dirt
and the grime
from the past to the future throught
the space and time
traveling deep beneath the waves...
but oh Lord we've got to fight
with the thoughts in the head
with the dark and the light
no use to stop and stare
and if you don't know where you're going
any road will take you there!


tests and results:

first test:
to compare the quality, sizes and times needed,
i did some samples using trim.

Total Frames: 2001
Total Time : 00:01:07<---sample time used, trim(30000,32000)

all samples with cq 60:
full script........9.362.678,00:05:32
only merge.....9.978.867,00:05:15<--less size,more time
no filters.......11.624.519,00:04:08<--more size,less time
only stf.........10.322.693,00:04:25


second test:
searching the prediction for each case in ToK, i found using:
Total Frames: 56417
Total Time : 00:31:22<---sample time used, Trim(0,56416)
the CQ obtained for each case was applied to do the final samples,
see the values of this CQs in the final test!


final test:
after found the CQs for each case:
Total Frames: 3001
Total Time : 00:01:40<---sample time used, trim(30000,33000)

full script.....17.024.579,00:08:13,CQ : 68,400<--more size,more time
only merge..16.454.478,00:07:49,CQ : 63,340
no filters......16.244.254,00:06:11,CQ : 57,260<--less size,less time
only stf........16.802.096,00:06:36,CQ : 63,270


the results:
after burn the 4 samples from my final tests:
in my phillips 29 with only 2 "months of age",
in another phillips 14 with less than 1 year of use
and another sansung 20 with 5 years:

ALL SAMPLES SEEMS THE SAME,TEENY DIFFERENCES.
I DON'T KNOW WHAT IS BETTER!!!

no filters - 16.244.254,00:06:11,CQ : 57,260<---less size,less time
is better for me than
full script - 17.024.579,00:08:13,CQ : 68,400<---more size,more time
cos the image seems the same,is faster to encode and give less size.
AND ALL SAMPLES FROM THE SAME SOURCE

sorry my screams "caps lock" cos after 3 days doing samples and
Kwag was changed the values of STF in the middle of my tests...
i did all again with the new STF parameters
was really hard

i have pictures and a friend
will help me to post it here.
the pictures are:
4 bmp CQ60,
4 bmp CQ60 in another frame,
4 bmp from dif Cqs and dif scripts,
4 bmp from dif Cqs and dif scripts in another frame
in rar(2,52mb) and/or zip(3,61mb) to send.

the same 16 pictures in jpg format
in zip(298kb) to send.

if anyone wish, i send the pictures for mail, just pm me!
tomorrow i will post it here, only waiting my friend
cos i don't know how to post it.

please,
i'm waiting advices and commentarys from you all!
thanks in advance.
Reply With Quote
  #73  
05-24-2003, 05:24 PM
Wolfi Wolfi is offline
Free Member
 
Join Date: Oct 2002
Location: Sweden
Posts: 489
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via AIM to Wolfi Send a message via MSN to Wolfi Send a message via Yahoo to Wolfi
Does the newest script increase CQ by 5 for every one? CQ shrinks for me by 0.02 cq value if I use the newest script. I compared the oldest and the new one and cq actually gets smaller

ConvertToYUY2()
LegalClip()
unfilter(50,50)
GripCrop( width=544, height=480 )
GripSize(resizer="BicubicResize")
STMedianFilter(8, 32, 0, 0, 8, 32)
SpaceDust()
temporalsmoother(1,2)
mergechroma(blur(1.50))
mergeluma(blur(0.2))
DctFilter(1,1,1,1,1,.5,.5,0)
GripBorders()
Letterbox(0,0,32,32)
LegalClip()

What am I doing wrong, or am I doing anything wrong

BTW: Jorel great job

//Wolfi
Reply With Quote
  #74  
05-24-2003, 09:26 PM
ovg64 ovg64 is offline
Free Member
 
Join Date: Jan 2003
Location: Puerto Rico
Posts: 423
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via MSN to ovg64
@ Jorel

The reason for the filters r to clean up the video image and reduce the file size and of course encode time is much longer, if we could fit more in one CD we wouldn't worry to much about filter and everything would be a lot easier, but as it is we have to sacrified speed for quality and i guess we will keep using this filter until a smart person comes along and tell us how to get more compression, good quality, and faster encode all in one deal. keep experimenting who knows maybe it will be you.
Reply With Quote
  #75  
05-24-2003, 11:23 PM
dazedconfused dazedconfused is offline
Free Member
 
Join Date: Nov 2002
Posts: 316
Thanks: 0
Thanked 0 Times in 0 Posts
Hi Wolfi,

Quote:
Originally Posted by Wolfi
Does the newest script increase CQ by 5 for every one?
Of course not...every single movie/source that a person converts will be somewhat different due to the various characteristics of the movie.

Quote:
CQ shrinks for me by 0.02 cq value if I use the newest script. I compared the oldest and the new one and cq actually gets smaller
A CQ difference of 0.02 is negligible...I highly doubt if anyone's eyes could detect a CQ difference that tiny. Kwag lowered the STMedianFilter values a bit because some people were complaining about the picture quality with the higher values, so that is why you are getting a slightly lower CQ level than before. Don't worry so much about CQ levels. I would much rather have a slightly lower CQ level and a better overall picture than vice versa. Besides, with the current "Optimal Script", you can go as low as ~CQ50 (note: I said approximately since different people have different tastes) and you'll still have a fairly acceptable quality when viewed on most standard TV's. I usually try to stay at or above ~CQ54 while using the latest script, but that's just a general personal preference based on my own encoding experiences and tastes.

Quote:
What am I doing wrong, or am I doing anything wrong
I don't think you're doing anything wrong friend! But I think maybe you're just a bit too worried about CQ levels. If I were you, I would drop the SpaceDust and DctFilter filters (unless you're converting from a very poor blocky source or you're extremely desperate for a smaller filesize so you can reach your target CQ). Personally, I have found that using both STMedianFilter AND SpaceDust is just too much filtering and the picture quality suffers, so I would try to use one or the other (and I'd take STMedianFilter instead of SpaceDust since it's quicker and does a better job). The less filters, the better! (and faster!). Just my $0.02...

Happy encoding,
-d&c
Reply With Quote
  #76  
05-24-2003, 11:53 PM
jorel jorel is offline
Invalid Email / Banned / Spammer
 
Join Date: Aug 2002
Location: Brasil - MG - third stone from the sun
Posts: 5,570
Thanks: 0
Thanked 0 Times in 0 Posts
thanks for observations ovg64


i was waiting more observations from other members
cos was hard to do and write the results...
maybe is better wait a little more.

you wrote:
"...to clean up the video image and reduce the file size ...
if we could fit more in one CD we wouldn't worry to much about filter and everything would be a lot easier....
have to sacrified speed for space and i guess we will keep using this filter ...."

or i misunderstand your answer or my explanation was unclear.
see:
if i don't see extreme differences in quality and got
CQ:57,260 without filters and CQ:68,400 in the full new script
after find the CQs for the samples after prediction,
we will got the same final size for them too.
if we encode the full movie after find the prediction,
no matter what filter we choose,the final size will be the same or near
my observations in the tests are:
352x240,we need some filters but, for 480x480 we don't need the same filters cos more "resize" gives less problems
and for 528x480, less filters than to 480x480.

resuming:
do a sample(same source) with CQ 57 and another with CQ68...
the differences is not too huge(only my opinion)...for me,
more resize means 528x480 that is more than 352x240
and give better quality with less filters.
and don't matter the resolution, the prediction will find
the ideal CQ with or without the filters.
the final size will be near in all cases.
one with more CQ and another with less CQ.

the important is not only the CQ or time to encode.
the best is the quality and i see only teeny diferences in 480x480
using only unfilter(20,20) or no one filter.
my sources are DVDs only, i don't encode captured avis or divx!
maybe this make a big difference!
i see in forums everybody changing filters,removing filters,
put a new one filters but, they don't got our results with kvcds.
i think that the miracle here is the kvcd formats and
the Kwags modifications for tmpgenc, not the filters.

or it all is my big mistake....
but my samples and pictures show me that i'm not too wrong!

thanks my friend ovg64, if my explanations here worse the situation and
turn it all more confuse,
or are not clear, please excuse me.
you know my horrible english!
Reply With Quote
  #77  
05-25-2003, 10:42 AM
ovg64 ovg64 is offline
Free Member
 
Join Date: Jan 2003
Location: Puerto Rico
Posts: 423
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via MSN to ovg64
Hey Jorel, say you have to do a movie an you determine that 60% CQ is acceptable for it, now get tok and tok tells you that the hole movie will fit at 55% CQ. Now you want to get up to 60% again and that is when the filter come in affect, now you put a couple of this filters and bang you are back to 60% or more CQ. Now i agree with you that the least filters we use the better, cause if you use too many filters you probably make things worse (image quality) not to mantion your encode is slower, so you use the least amount of filters with the best posible setting depending on the movie. Personally I thinck anithing encode above 75% or even 70% you dont need anything but resizeing filter I guess Kwag templates are that good.
Reply With Quote
  #78  
05-25-2003, 11:14 AM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by ovg64
I guess Kwag templates are that good.
The secret is not in the templates, but in the matrix

-kwag
Reply With Quote
  #79  
05-25-2003, 02:44 PM
Kane Kane is offline
Free Member
 
Join Date: Jan 2003
Posts: 398
Thanks: 0
Thanked 0 Times in 0 Posts
ähhmmm....i´m a little bit confused, reading your last posting jorel:

Quote:
if i don't see extreme differences in quality and got
CQ:57,260 without filters and CQ:68,400 in the full new script
after find the CQs for the samples after prediction,
we will got the same final size for them too.
ok, i called this threat Death of STMedianFilter, but did i get you right?
YOU KILLED ALL FILTERS
__________________
greetz Kane
Reply With Quote
  #80  
05-25-2003, 04:37 PM
jorel jorel is offline
Invalid Email / Banned / Spammer
 
Join Date: Aug 2002
Location: Brasil - MG - third stone from the sun
Posts: 5,570
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by kwag
Quote:
Originally Posted by ovg64
I guess Kwag templates are that good.
The secret is not in the templates, but in the matrix

-kwag
Kwag and all friends,
i post in my biiig explanation(well, was trying...):

"i think that the miracle here is the kvcd formats and
the Kwags modifications for tmpgenc, not the filters."

my source have more or less 82minutes(dvd) and i got big results as
posted and all my samples show me exact what i wrote.
i send my pictures to ovg64 compare and he can
explain better than me cos he speaks great english and
his observations will be really true, of course


@helper ovg64,
as i posted my friend, your opinion is very important.
you have the pictures to compare.

@ big tester Kane,
Kwag is the fastest big clicker in the world,
you hurt STF with mortal "punch" and
i am the bad and worse dangerous filters killer!


ps:
maybe kwag and more friends are thinking now:
this brasilian mid old guy(more old than mid)is mad.

only kidding...!

but the results are here,if someone want it just pm!

thank you all!
Reply With Quote
Reply




Similar Threads
Thread Thread Starter Forum Replies Last Post
Avisynth: STMedianFilter slowdown? scrappy Avisynth Scripting 9 01-31-2006 01:41 PM
An Irish death glänzend Off-topic Lounge 0 09-04-2004 05:37 AM
Avisynth: STMedianFilter in the Latest Scripts For VHS and captures? dongxu Avisynth Scripting 3 02-26-2004 09:51 AM
Avisynth: STMedianFilter v 0.1.1 (YV12 bug fix) jorel Avisynth Scripting 0 06-22-2003 07:00 AM
Avisynth: StMedianFilter + TemporalSoften Dialhot Avisynth Scripting 2 06-19-2003 09:38 AM

Thread Tools



 
All times are GMT -5. The time now is 05:37 PM  —  vBulletin © Jelsoft Enterprises Ltd