Avisynth: Death of Stmedianfilter - Page 5

jorel · #81 05-26-2003, 03:20 AM

me again friends.

after a big search i found some importants posts
that "talk" about the order of the filters in the script and
what happen if his order is changed.

the resume is:

if resize is before the filters,
you got more final size, less time to encode

if resize is after the filter,
you got less final size, more time to encode

this is not "the law", but the order change how tmpgenc will work.
just like unfilter used before resize,
every script need tests with samples to show the
quality,size and time in the results.

http://www.kvcd.net/forum/viewtopic....b8258d0c0634d0

http://www.kvcd.net/forum/viewtopic....b8258d0c0634d0

http://www.kvcd.net/forum/viewtopic....b8258d0c0634d0

and the last that i found:
http://www.kvcd.net/forum/viewtopic.php?t=3790&start=16

Jellygoose · #82 05-26-2003, 10:52 AM

hi guys...

wow, everything is getting so confusing right now in my eyes... well the current optimal script is really good, but seriously, I sometimes think that I already did better encodes, some months ago, when I watch a movie. Is that just me?

well maybe it is... I'm waiting to see those pictures from Jorel... it's hard to believe, that a script with no filters looks as good as the current script... however, there are so many wonders in the world of video conversion

Kane · #83 05-26-2003, 01:37 PM

hi jorel

i´m quite impressed, the are nearly NO DIFFERENCES between the script with and without filters

i can scroll through all pictures actually, they have all the same quality

but i can´t believe it....

jorel · #84 05-27-2003, 02:10 AM

Quote:

Originally Posted by Kane

hi jorel

i´m quite impressed, the are nearly NO DIFFERENCES between the script with and without filters

i can scroll through all pictures actually, they have all the same quality

but i can´t believe it....

yes Kane,

this is what i'm trying to show.

i did this 16 images from the samples and
if the images was not named to send,
you will never know how is with one filter
and how is without the filters.

but you can see that each image have little differents in kb size

Jellygoose

i'm trying to send this images for you but
"don't know why",your mail (or mine), don't work!
the address is correct in pm?

Kane · #85 05-28-2003, 05:03 PM

finally i made some tests too with 4 script, prediction with ToK, 2min sample 528x576

1.

Code:

my current script
mergechroma/luma, temporalsoften, unfilter before resizing
Encoding... CQ : 70,608
time: 00:08:07

2.

Code:

no filters!!
Encoding... CQ : 64,642
time: 00:05:30

3.

Code:

mergechroma/luma , temporalsoften, sharpen before resizing
Encoding... CQ : 70,010  8O
time: 00:10:08

4.

Code:

mergechroma/luma , temporalsoften , sharpen after resizing
Encoding... CQ : 64,672  8O
time: 00:08:59

jorel, you are right. there´s no difference to see between script 1 and 2 if you are looking at still images, but in my samples the sample with filters (script 1) was visibly sharper.

script 4 was the looser of the competition. quite fast, but bad quality

script 3 gave me a bit more sharpness, hardly visible but there was an increase of sharpness, cq values of 1 and 3 are nearly identical, but sharpen(0.5) is really slow

it took two minutes

more to encode a 120sec sample, this would be
100minutes more for a 100minute movie

hmmm. not sure whether to replace unfilter by sharpen in my script
let´s see, what the final encoded movie looks like

jorel · #86 05-28-2003, 05:37 PM

friend Kane

your results are precious!

see what i'm testing :

Legalclip()
sharpen(0.5)
BlockBuster(method="noise", detail_min=3, detail_max=10, variance=1.0, seed=0)
BicubicResize(448,448,0,0.6,0,0,720,480)
AddBorders(16,16,16,16)
Legalclip()

this script for me is faster,with few less size and great quality.
sharpen instead unfilter give more sharpness and less size
and blockbuster give less noises.

is only for avisynth 208 and i'm reading the thread about
avisynth 251 but i'm confused and waiting for Kwag and
all friends(you include) got final results.

thank you very much!