Low Bitrates - compressibility techniques

[bilu]

Hi,

Imagine you would want to fit 90 minutes of progressive PAL in 560 MB (it's about 860 kbps), high-action, dark, noisy, whatever you feel it's hard to compress.

My results so far
=================

At 0.2 bpp (512x320 @860kbps) I haven't seen good results, haven't tested much with 0.25 though.

Best results I got until now (testing with Gladiator Chapter 15 and others) with

Deen("a2d",2,10,12)
TemporalCleaner(5,10)
Undot
Unfilter(-5,-5)

Still have hopes for DCTFilterD (see thread "Integer DCT/iDCT filter questions" (http://forum.doom9.org/showthread.php?s=&postid=268620#post268620) ).
DCTFilterD(6) seems the best to me but as you can see in that thread it's not improving compressibility.

Mpeg2dec3 luma deringing blurs too much and chroma deringing "greens" everything, worst than H263. But deblocking seems OK.

Tried Conv3d(movieHQ and movieLQ),deen("a3d",1,10,12),Fluxsmooth(5,5) and mpeg2dec3 PP. Haven't tried Dust.

Torches on dungeons or caves also seem to affect encoding, specially with B-frames.

Testing with latest Koepi's XVID builds and XVID stable, following Iago's guides, and FFVFW (libavcodec is somewhat mature but this codec wrapup still needs to grow stable, looks better than XVID @ low bitrates though)

==================

And now YOU :D

What kind of approach would you use to increase compressibility, both on filtering and resolutions?


Bilu

[SILICON]

Try convolution3D, TemporalSoften and CNR2.
Work well for me.

[Manao]

For me, usually :

* crop vertically
* fluxsmooth ( 2 - 10 , 2 - 10 ) depends of the movie
* crop horizontally
* resize ( lanczos or bilinear, depends of the movie, the resolution, and the space I have )
* convolution3d( something between movieHQ and movieBQ, sometimes more for very old films ( B & W very grainy, like "bringing out baby" ) )
* temporalsoften( the 'conservative' settings given by Shodan : 3,3,3,10,2 or something like that )
* DCTFilter(1,1,1,1,1,1,0.5,0)

I know it's a lot of filtering, but I want the movie on one CD, with two soundtracks ( ogg@q2.5 + xvid + b-frames ), with a decent resolution ( min 512*384 or 576*320 or 640*272, depends of the AR )

For your case, if it's only 90 minutes, I think you can also save some space on the soundtracks to gain 30 Mo for the video.

Finally, if you plan to use b-frames with xvid, denoise a lot to avoid flickering in the background.

@SILICON : why cnr2 ? It's made for VHS / DV ripp mainly, no use for a normal DVD ripp.

[bilu]

Well... It's time to post my full script here :)

#=======================================PLUGINS================

Loadplugin("D:\avs_filters\avs25\AutoCrop.dll")
Loadplugin("D:\avs_filters\avs25\avsmon25a.dll")
Loadplugin("D:\avs_filters\avs25\Decomb.dll")
Loadplugin("D:\avs_filters\avs25\MPEG2Dec3.dll")
Loadplugin("D:\avs_filters\avs25\VSFilter.dll")
#Loadplugin("D:\avs_filters\avs25\UnDot.dll")
Loadplugin("D:\avs_filters\avs25\Deen.dll")
Loadplugin("D:\avs_filters\avs25\UnFilter.dll")
Loadplugin("D:\avs_filters\avs25\BicublinResize.dll")
#Loadplugin("D:\avs_filters\avs25\DctFilter.dll")
Loadplugin("D:\avs_filters\avs25\TemporalCleaner.dll")
Loadplugin("D:\avs_filters\avs25\FluxSmooth-2.5.dll")
#Loadplugin("D:\avs_filters\avs25\Convolution3DYV12.dll")
#Loadplugin("D:\avs_filters\avs25\blockbuster.dll")

#=======================================FUNCTIONS========

function ResizeARC(clip c,int bitrate,float compressibility,string resizefn) {
ratio=string(eval(string(c.width)+". /"+string(c.height)+"."))
res=string(floor(eval("("+string(bitrate)+"*1000)/("+string(compressibility)+"*"+string(c.framerate)+")")))
nhtemp=floor(sqrt(eval(res+"/"+ratio)))
nwtemp=floor(eval(string(nhtemp)+"*"+ratio))
nh=nhtemp - nhtemp%32
nw=nwtemp - nwtemp%32
return eval(resizefn+"(c,"+string(nw)+","+string(nh)+")")
}

function AutoIVTC(clip c){
IVTC= (c.framerate > 25)? "Decimate(Telecide(c),cycle=5)":"FieldDeinterlace(c)"
return eval(IVTC)
}
#=======================================SOURCES=====================

mpeg2source("D:\GLADIATOR\chap15\chap15.d2v",idct=2,cpu=0)

#=======================================IVTC========================

#AutoIVTC()

# Wrote just for fun :)

#=======================================CROPPING===================

#YV12toYUY2()
#AutoCrop(mode=1,threshold=40,samples=10,hMultOf=8,wMultOf=16)
#ConvertToYV12()

# AutoCrop is time-consuming, it requires YUY2, but gives the crop values in Preview mode :)

#chap15
crop(4,76,712,432)

#=======================================SUBTITLES=========

#SwapUV()
#vobsub("D:\AND_NOW_FOR_SOMETHING_COMPLETE\VIDEO_TS\VTS_01_0")
#SwapUV()

#SwapUV is required with Vobsub to get the right color in permanent subtitles

#=======================================FILTERING/RESIZING======

ResizeARC(860,0.2,"FastBilinearResize")

# My first function ;) It respects AR as possible mantaining MOD32 resolutions,uses bitrate, bpp and the resize function specified as parameters (to newbies, see doom9 divx5 gknot encoding guide for explanation on bpp)

FluxSmooth(5,5)
TemporalCleaner (ythresh=5, cthresh=10)
deen("a3d",1,10,12)
Unfilter(-5,-5)
LumaFilter(+15,1.2)

#=======================================FINAL PROCESSING=====

Limiter()
MonitorFilter()


Got the best result ever with this denoising sequence. XVID MPEG quant and DIVX5 B-frame setting (1/150/100) - tested the 1st 2400 frames of chapter 15 on Gladiator for 10080 KB.


I hope my resize function pleases some of you :p


Bilu

[The Link]

@ bilu

Just wanted to say that your script would be worth being included in avisynth.org in my eyes. :)

Everything I´ve ever tried regarding filters for better compressibility is represented in this thread (and some more ;) ) so theres nothing I could contribute. :(

Regards,

The Link

[frank]

Gladiator on 1 CD?? What a crap!
Gladiator has about 150 min length, and not 90.
This movie is an example of good encoded material and sound.
Do you want it to press and cut until it looks blurred ugly?

You must learn:
High-action, sharpness -> much information.
Much information -> more encoded bits! You can watch that on a bitrate viewer showing VBR encoded material.

Every filter means LOST of information. There are no wonders.

Better use 99min/90min CDs and take 2.

[bilu]

@frank

I'm not encoding Gladiator on 1 CD :rolleyes: just some hard-to-compress chapters at the same quality I would go for 90 minutes on a 650 MB CD.

It's not about miracles, just worst-case testing :o


Bilu


PS: By the way, I'll try to put my resize function on avisynth.org :p

[bilu]

@sh0dan

WOW :D It's great to share knowledge like this. Viva Wiki !!!

@all

http://www.avisynth.org/index.php?page=ShareFunctions

see "Resize based on bits/pixel*frame formula (bpp)"



Best regards,

Bilu

[SILICON]

Originally posted by Manao
why cnr2 ? It's made for VHS / DV ripp mainly, no use for a normal DVD ripp.

CNR2 blurr the croma and it`s very fast.

I don't see any chage in the image quality, and the needed bitrate is lower.

Try it.

I use:
LoadPlugin("C:\DVD2CVCD\PLUGINS\Convolution3dYV12.dll")
LoadPlugin("C:\DVD2CVCD\PLUGINS\CNR2YV12.dll")
LoadPlugin("C:\DVD2CVCD\PLUGINS\MPEG2DEC3YV12.dll")
# the deblock reduces the size, and blur very litle
Mpeg2Source("C:\DVD2CVCD\Pelicula\PELICULA.D2V", IDCT=2, CPU=4, iPP=FALSE)

# Cut the first trailers of the film.
TRIM(30,184875)

# Erase small noise in the black bars. Speed the codification.
CROP(14,74,692,434)

# less blocks in the dark areas
LumaFilter()

# clear the dots in some films
UnDot()

# the resize before, for speed.
BilinearResize(336,318)
Convolution3D(0, 3, 4, 3, 4, 3, 0)
CNR2()

# the croma is filter by CNR2
TemporalSoften(1,5,0,7,2)

# Restore the 4/3 aspectratio (CVD PAL format)
AddBorders(8,129,8,129)

# Legalcy values. less size ????
Limiter()

# To send the MPEG2 compressor
ConvertToYUY2()

¿too many filters?

[jonny]

bits/pixel*frame it's a bad parameter for choose the resolution.

It's far better to make a real compressibility test to decide the resolution.

[bilu]

@jonny

This formula was allways an option for those of us who choose automization over customization, and I think having options is allways good.

Some people like Windows, others like Linux. Some like Flask,DVDx or Vidomi, others Avisynth and XVID, others yet libavcodec and transcode. :rolleyes:

I've seen recently a poll on this forum under Divx5 or GKnot (don't remember) wondering if people found the compressibility test useful, and I've seen a lot of mixed answers.

Personally I think it's the best way, but I have the option to avoid it if I choose a less perfect encode over a compressibility test for each movie.


Bilu

[jonny]

This formula was allways an option for those of us who choose automization over customization, and I think having options is allways good.

I agree, and the script it's amazing, really original.

I've seen recently a poll on this forum under Divx5 or GKnot (don't remember) wondering if people found the compressibility test useful, and I've seen a lot of mixed answers.

http://forum.doom9.org/showthread.php?s=&threadid=42492
(the answers are not so mixed ;))

Cheers
jonny

[Prettz]

If you want a worst-case scenario I suggest you get the Aliens DVD. The whole DVD looks like it's being output from a 1 CD divx3 rip. I've never seen so much quantization artifacting on a DVD in my life.

[bilu]

@prettz

I got that DVD but I borrowed it too long ago, don't know if I'll ever get my hands on it again... :stupid:
But you're right, a very good source with chroma headaches ;)

Some chapters from the Gladiator are very good too: the battle in Germania and the first fight in the Coliseum, for example.

@jonny

I share N_F's point of view on that thread. It is possible if you are an experienced encoder to look for the kind of stuff that affects compressibility and try to get rid of it as possible, not necessarily sacrificing detail. To an extent, you can do the compressibility test with your own eyes. :)

I've been making tests again with Chapter 15 from the Gladiator (7:30)
trying to squeeze it to 47250 KB (again the average 860 kbps), using latest umaniac build, VHQ=1, B-F=1/150/100, H263 quant, no CM, no CO,
no Qpel.

My current denoising sequence (this chapter as a lot of smoke):

ResizeARC(860,0.2,"BicubicResize")
deen("m2d",1,10,12)
Undot()
Unfilter(-5,-5)
LumaFilter(+16,1.1)

deen("m2d") seems really good gathering the smoke, and it's good to reduce the motion vectors used as much as possible for low bitrate (while trying to preserve detail of course).

Bilu

[jonny]

If you share the point of view of N_F, why are you using 0.2 (that it's a worst indicator compared to comp.test)?
I agree that if you are experienced, you can make a good valutation without comp.test, but at this point, using 0.2 seems to me conflicting

[bilu]

Yes, I know ;) but I still think that very good denoising (like deen(m2d)) can do it. This were my sources for values:


http://www.doom9.org/xvid.htm
=============================

It's a bit guesswork as you cannot predict compressibility of your movie but here's some hints:

* It's not a good idea to go below 0.20 bits/pixel*frame
* If you go above ~0.35 you might not be able to reach your desired filesize
* If you go for 1 CD stay close to 0.20
* If you go for 2 CDs stay around 0.27
* If TV bits/pixel*frame is below ~0.13 you should raise your bitrate (and go for more CDs).

Mode the slider until you're within this range. These values are the general rule for DivX3 encoding. I find it safe to go quite a bit lower using another codec, say 10-20% at least.


http://www.doom9.org/gknot-basics-divx3.htm
===========================================

It's a bit guesswork as you cannot predict compressibility of your movie but here's some hints:

* It's not a good idea to go below 0.20 bits/pixel*frame
* If you go above ~0.35 yo might not be able to reach your desired filesize
* If you go for 1 CD stay close to 0.20
* If you go for 2 CDs stay around 0.27
* If TV bits/pixel*frame is below ~0.13 you should raise your bitrate (and go for more CDs).

Mode the slider until you're within this range. These values are the general rule for DivX3 encoding but you should not consider them to be the laws of nature. I've encoded many movies that had a lower value and they still looked great.


http://www.doom9.org/gknot-basics-divx5.htm
===========================================

It's a bit guesswork as you cannot predict compressibility of your movie but here's some hints:

* It's not a good idea to go much below 0.17 bits/pixel*frame
* If you go for 1 CD stay close to 0.17
* If you go for 2 CDs stay around 0.25
* If TV bits/pixel*frame is below ~0.13 you should raise your bitrate (and go for more CDs).

Mode the slider until you're within this range. These values are a rough estimate for DivX5 encoding but you should not consider them to be the laws of nature. Sometimes values below the given ones may result in a shitty looking movie and in other cases going below these values will still result in a great looking movie.

http://www.mplayerhq.hu/DOCS/tech/encoding-tips.txt
===================================================

Now if you want to decide which resolution (and scaling parameters)
to chose you can do that by looking at the $bpp:

< 0.10: don't do it. Please. I beg you!
< 0.15: It will look bad.
< 0.20: You will notice blocks, but it will look ok.
< 0.25: It will look really good.
> 0.25: It won't really improve visually.
> 0.30: Don't do that either - try a bigger resolution instead.

Of course these values are not absolutes! For movies with really lots
of black areas 0.15 may look very good. Action movies with only high
motion scenes on the other hand may not look perfect at 0.25. But these
values give you a great idea about which resolution to chose.


What do you suggest? :)

Best regards,

Bilu

[jonny]

I suggest to read all the guide:

http://www.doom9.org/gknot-divx5.htm

And last but not least the compressibility check.

It's possible to run a short first pass to find out the optimal compressibility of a movie (and make an even better resolution choice). Turn the Compressibility check on and set it to 5%, then press the Now button, wait for a few minutes until GKnot comes up again. The Edit button allows you to edit the AviSynth script prior to performing the check but only experienced people should touch that.


The compressibility check gives you a rough indication of how good a movie can be compressed. The 3 values shown above are related as follows: the 0.338 is the value that the compressibility check returned. If the Bits/(Pixel*Frame) the first value, right above the Load button matches this value you have perfect quality, going for a higher value would be overkill as the movie wouldn't look any better.. The limit of the codec has been reached. The 41.7% value tells you the relationship between the optimal case and what you're going to get at the currently selected resolution. 60% or higher will pretty much guarantee that you'll get pretty good looking results. Going over 80% is a serious overkill and you should rather increase the resolution. Values in the range of 50-60% still look pretty good and 40-50 is still OK, especially when you activate B-frames. The darker a movie gets and the cleaner the source the more you can compress a movie. Using B-frames certain movies will look good even at a 0.12 Bits/(Pixel*Frame value). Once the value is in the acceptable range set the trim option to Both (enc. credits separately).

[bilu]

I don't know if this is the correct interpretation, but here it goes:

The 3 values shown above are related as follows: the 0.338 is the value that the compressibility check returned. If the Bits/(Pixel*Frame) the first value, right above the Load button matches this value you have perfect quality, going for a higher value would be overkill as the movie wouldn't look any better.The limit of the codec has been reached.

Codec max=0.338

The 41.7% value tells you the relationship between the optimal case and what you're going to get at the currently selected resolution.

0.338 * 0.417 = 0,140946

60% or higher will pretty much guarantee that you'll get pretty good looking results.

0.338 * 0.6 = 0,2028

Going over 80% is a serious overkill and you should rather increase the resolution.

0.338 * 0.8 = 0,2704

Values in the range of 50-60% still look pretty good and 40-50 is still OK, especially when you activate B-frames.

50% = 0,169
60% = 0,2028
40% = 0,1352


It seems to make sense with what I've read too, if my interpretation is correct, that is.


Bilu

[bilu]

Deen("m2d") and edeen are not stable filters, and Marc FD has left the building (sure hope he comes back :rolleyes: ).

I never explored C3D beyond the presets until today.

Convolution3D (1, 6, 10, 6, 8, 2.8, 0)
ResizeARC(860,0.2,"BicubicResize")
Undot()
Unfilter(-5,-5)
LumaFilter(+16,1.1)

I have done a good looking rip of Chapter 15 with it, preserving much more detail than with any of my previous tests and still looking better!
:D

@trbarry
The effect of this filtering reminds me of DctFilterD(7) but without the ringing and "cheating" on motion vector detection. I'm serious, have a look, this can be what you were looking on this thread (http://forum.doom9.org/showthread.php?s=&threadid=38539)

Best regards,
Bilu

[jonny]

0.338 is only an example :)

I'll try to be clean.
bpp tell you almost nothing about the final quality.
there will be cases when 0.2 will be ok, cases when will be too much, cases when will be too low. The same goes for 0.1 or 0.3, it's not a good indicator.
I can elaborate this in a different way: using 0.2 you'll obtain a different average quantizer from movie to movie, because every movie compress in a different way, even if you are using the same bpp.

Comp.test it's a good indicator, for example, if you obtain something like 80%, you can be sure that the average quant will be something like 3 and so on.
Comp.test compare the size of the video encoded at max quality, with the size you are targeting, this indicator is POWERFUL.

Now i stop here my promotion about the comp.test :D
If you like bpp, stay with bpp.

[bilu]

I think I haven't got that much experience on comp. tests to feel so sure :)

Mainly I've been improving my denoising techniques on high-action noisy sequences. But there are different kinds of noise and there is a possibility that motion itself after normal denoising is too high (try the one in this link (http://www.tecoltd.com/enctest/enctest.htm) for example).

But still you convinced me to check more about comp. test and if denoising is enough for low-bitrate encoding.
Be happy man! :D


Bilu

[jonny]

But still you convinced me to check more about comp. test and if denoising is enough for low-bitrate encoding.
Be happy man!

Nice choice :)

[yaz]

Originally posted by jonny

bpp tell you almost nothing about the final quality ... Comp.test it's a good indicator ... Comp.test compare the size of the video encoded at max quality, with the size you are targeting, this indicator is POWERFUL.


i think i got what u mean & u're right in a way but ...
if compressibility is a good indicator then bpp can't be wrong. simply, because they're interrelated. compression, as given by gknot, is the rate of the bpp set before 1st pass (bpp0) & that obtained afterward (effective bpp). check it out! as this, it's as good as compression (i prefer this term, as 'compressibility' has a taste of a kinda abs value:-)

u're right; bpp0 does not guarantee anything about quality but it's also true for compression. again, it's true that the higher the compression value is (the lower the effective compression!) the lesser details are removed & so the closer is the final quality (whatever it means) to the original. it's all true but ... what to do with it? after all, u must check the results 'visually'. i mean, that's the best which u like the most :-)

about compressibility ckeck. it works in general but i was misled by some times. consider what it does! it checks 5% in a way of selecting the first 14 frames outta each 280. this way has some drawbacks. e.g. it generates lottsa fake keyframes. fake keyframes generate fake delta frames ... :-) imho, the most reliable compression check is the 1st pass itself. i don't state comp.check is wrong(!) i just say its use is a bit risky.

the bests
y
ps. kick me if i was wrong :-)

[jonny]

i think i got what u mean & u're right in a way but ...
if compressibility is a good indicator then bpp can't be wrong. simply, because they're interrelated. compression, as given by gknot, is the rate of the bpp set before 1st pass (bpp0) & that obtained afterward (effective bpp). check it out! as this, it's as good as compression (i prefer this term, as 'compressibility' has a taste of a kinda abs value:-)

current_bpp / 1st_pass_bpp
with some math, simplify to:
= target_size / predicted_size_qb100%

As you can see bpp alone is quite unuseful, the heart of the test is compare target size with the size at qb 100% (or bitrate / bitrate_to_obtain_qb100%_size).

You can read this thread:
http://forum.doom9.org/showthread.php?s=&threadid=44621
(contains some math, i think it's useful to understand what i say above... note that this thread mainly refer to the manual method here: http://forum.doom9.org/showthread.php?s=&threadid=44414)


u're right; bpp0 does not guarantee anything about quality but it's also true for compression. again, it's true that the higher the compression value is (the lower the effective compression!) the lesser details are removed & so the closer is the final quality (whatever it means) to the original. it's all true but ... what to do with it? after all, u must check the results 'visually'. i mean, that's the best which u like the most :-)

If you encode at 80% comp.test value, you are guarantided to have high quality.
If you encode at 30% comp.test value, you are guarantided to have poor quality.
I can agree that the comp.test isn't the only value to take care, but is one of the most powerful toys we have to adjust the quality.



about compressibility ckeck. it works in general but i was misled by some times. consider what it does! it checks 5% in a way of selecting the first 14 frames outta each 280. this way has some drawbacks. e.g. it generates lottsa fake keyframes. fake keyframes generate fake delta frames ... :-) imho, the most reliable compression check is the 1st pass itself. i don't state comp.check is wrong(!) i just say its use is a bit risky.

Fake keyframes are discarded in comp.test (with b-frames, even some oversized b-frames are discarded... (DivX5Enc or GKnot do this) there is a lot of work behind... trust me).
How accurate is the method?
Check those pages on my site:
http://jonny.leffe.dnsalias.com/whywork.htm
http://jonny.leffe.dnsalias.com/sptests.htm

EDIT:

fake keyframes generate fake delta frames


This is not true

[yaz]

oh geee .. i didn't want to harm u. in general, we talk about the same. in essence; wout a comp.check, proper encoding is a quest of luck. isn't it?

Originally posted by jonny
current_bpp / 1st_pass_bpp
with some math, simplify to:
= target_size / predicted_size_qb100%

this equation is good if current_bpp is what u set before 1st pass & 1st_pass_bpp is what u got after that. otherwise, it's wrong. wout any further math :-)))

Originally posted by jonny
As you can see bpp alone is quite unuseful, the heart of the test is compare target size with the size at qb 100% (or bitrate / bitrate_to_obtain_qb100%_size).

- what do u mean 'bpp alone' ? we've got 2!
- anyway, just put your infos together. the rate u want is the rate of the bpps.

Originally posted by jonny You can read this thread: ... [/B]

i know your worx on this field well & i highly appreciate it. really.

Originally posted by jonny Fake keyframes ...[/B]

maybe, i used a wrong term here. i'm not a native, so my english is a crap. i meant 'fake' that's the position of the keyframe in the stream is wrong. on encoding there wont be any keyframe there. not more.
anyway, not that's my main problem with 'partial comp.checks'

the bests
y

ps sorry again, if u felt me 'harming'

[jonny]

I usually write "in hurry" and i know that sometime my post seems rude, isn't my intention.

>oh geee .. i didn't want to harm u. in general, we talk about the same. in essence; wout a comp.check, proper encoding is a quest of luck. isn't it?

Yes, you are completely right.

>this equation is good if current_bpp is what u set before 1st pass & 1st_pass_bpp is what u got after that.

You are right again.

>- what do u mean 'bpp alone' ? we've got 2!
>- anyway, just put your infos together. the rate u want is the rate of the bpps.

I've written "bpp alone" mainly to make clear the concepts i've written in my previous posts here (for bilu), to make clear that bpp alone (before making a comp.test) isn't a good indicator (you already know this, so sorry :))

My english is crap too, i'm italian, do you speak italian? :)

[bilu]

I finally started doing some objective analisys with Moonwalker's XVID Analizer and DebugView (I'm using B-frames) on the compressibility subject.

The tests I've made already lead me to one conclusion:

- High-action and low bitrates only mix with extreme denoising/blurring
and very low resolutions :rolleyes:.Denoising is not necessarily enough on high-action sequences.

Yes, I know, you can all say "I told you so" :devil: , but I was looking for limits and learning on my mistakes ;)

Originally posted by frank
Gladiator on 1 CD?? What a crap!
Gladiator has about 150 min length, and not 90.
This movie is an example of good encoded material and sound.
Do you want it to press and cut until it looks blurred ugly?

You must learn:
High-action, sharpness -> much information.
Much information -> more encoded bits! You can watch that on a bitrate viewer showing VBR encoded material.

Every filter means LOST of information. There are no wonders.

Better use 99min/90min CDs and take 2.

Yes, Frank, you were right. That's part of what I learned by taking some risks. Sometimes you learn better that way than just "reading the other's Bible", if you know what I mean ;)


And now, to simplify your compressibility checks:

function Comptest(clip c, int percent) {
frange=floor(eval(string(c.framecount)+".*("+string(percent)+"./100.)"))
return eval("SelectRangeEvery(c,"+string(frange)+",14)")
}


Usage: Comptest(5)
For a compressibility test on 5% of the movie.


Best regards,
Bilu

[bilu]

Diary of a Madman part 9999....

Just to finish this thread, I was able to get nicer results (compressibility 60% for 5% movie) with an average bitrate of 1080 kbps from Gladiator Chapter 15.

The settings?

Avisynth 2.5
============
Mpeg2dec3 cpu=6
Bicubic 512,320
Deen("a3d")
Unfilter(-5,-5)

XVID 08032003 - Koepi's
=======================
H.263
3 b-frames, ratio 150, offset 100, threshold 0
VHQ=4
Chroma motion
Precision 6

Still it's not really decent... :sly: not something you'd like to watch and show to your friends.
So I'm finished with the Low-Bitrate tests! :D


This will be my base script from now on. I hope this thread was useful for some of you, especially for the newbies ;)



#=======================================PLUGINS=======================================

Loadplugin("D:\avs_filters\avs25\AutoCrop.dll")
Loadplugin("D:\avs_filters\avs25\avsmon25a.dll")
Loadplugin("D:\avs_filters\avs25\Decomb.dll")
Loadplugin("D:\avs_filters\avs25\MPEG2Dec3.dll")
Loadplugin("D:\avs_filters\avs25\VSFilter.dll")
Loadplugin("D:\avs_filters\avs25\UnFilter.dll")
Loadplugin("D:\avs_filters\avs25\CALL_25.dll")
Loadplugin("D:\avs_filters\avs25\Deen.dll")

#=======================================FUNCTIONS=======================================

function AutoIVTC(clip c){
IVTC= (c.framerate > 25)? "Decimate(Telecide(c),cycle=5)":"FieldDeinterlace(c)"
return eval(IVTC)
}

function ResizeARW(clip c,int nwtemp,string resizefn) {
nhtemp=floor((c.height * nwtemp)/c.width)
nh=nhtemp - nhtemp%16
nw=nwtemp - nwtemp%16
return eval(resizefn+"(c,"+string(nw)+","+string(nh)+")")
}

function Comptest(clip c, int percent) {
frange=(c.framecount*percent)/100
return SelectRangeEvery(c,frange,14)
}


#=======================================SOURCES=======================================

#mpeg2source("D:\AND_NOW_FOR_SOMETHING_COMPLETE\VIDEO_TS\monty.d2v")

#mpeg2source("D:\ANNA_AND_THE_KING\VIDEO_TS\chap2.d2v",idct=2,cpu=0)
#trim(2100,3300)
#last=trim(520,750)+trim(6150,6200)

mpeg2source("D:\GLADIATOR\chap15\chap15.d2v",idct=2,cpu=0)
#trim(1900,3600)
#trim(0,2400)

#mpeg2source("D:\star_trek\star.d2v",idct=2,cpu=6,moderate_h=40,moderate_v=40)

#AVISource("D:\avs_filters\avs25\tests\analyse\test.avi",false)
#AVISource("d:\test_clip\test.avi",false)
#ConvertToYV12()

#=======================================IVTC===========================================

#AutoIVTC()

#=======================================CROPPING=======================================

Autocrop(mode=0,wMultOf=16,hMultOf=16,threshold=40)

#=======================================COMPRESSIBILITY==============================

#Comptest(5)

#=======================================SUBTITLES=======================================

#vobsub("D:\AND_NOW_FOR_SOMETHING_COMPLETE\VIDEO_TS\VTS_01_0")

#=======================================FILTERING/RESIZING==============================

ResizeARW(512,"BicubicResize")
Temporalsoften(3,5,5,mode=2,scenechange=10)
deen("a2d",2,10,12)
Unfilter(-5,-5)
LumaFilter(+16,1.1)

#========================================FINALPROCESSING================================

#ShowFrameNumber(scroll=true)
Limiter()
MonitorFilter()



Best regards to all,
Bilu

[gilesw]

this is the only filter i can't track down. Anyone got a url ?

I don't even know what it does .

G.

[bilu]

CALL command:

http://forum.doom9.org/showthread.php?s=&threadid=46506

Best regards,
Bilu