Need help to improve PQ and compressability of Stargate SG1

[tetsuo55]

Hey everyone,

I recently got the original PAL boxed set of Stargate SG1.

Starting with the first season i noticed some problems with the source(i bet most of you have already seen these problems)

-The image is a bit on the soft side
-The episodes where filmed with 16mm film, and the grain is painfully visable
-I'm not sure if the grain is to blame or the low bitrate, but especially in the shadows there are visable flashing blocks
-Colors seem a bit strange on monitor, but look okay(still not great) on a crt tv(i want it to look good on monitor)


Now the grain and flashing blocks take up a lot of the bitrate, so i was hoping to do the following:

-Clean up some of the grain and flashing blocks so the image looks better
-Remove some of the grain to increase compressability
-Sharpen the image a bit to get rid of the softness
-Fix the colors
-The image is anamorphic but has some dark lines on all sides, should these be removed?

Currently i had the following in mind:

FFT3DFILTER(sigma=0.6, bw=14, bh=14, ow=7, oh=7, bt=2, dehalo=1, hr=1, sharpen=0)
grainoptimizer()
SeeSaw(NRlimit=0, NRlimit2=5, Sstr=2.22, Szp=1, Slimit=9, Sdamplo=16, Spower=1, sootheT=95, SdampHi=39)
ColorMatrix(Mode="Rec.601->Rec.709",clamp=false)


will this be enough?

I posted some screencaps (jpg compression makes it look a bit worse)

[Dark Shikari]

FFT3DFILTER(sigma=0.6, bw=14, bh=14, ow=7, oh=7, bt=2, dehalo=1, hr=1, sharpen=0)
grainoptimizer()
SeeSaw(NRlimit=0, NRlimit2=5, Sstr=2.22, Szp=1, Slimit=9, Sdamplo=16, Spower=1, sootheT=95, SdampHi=39)
ColorMatrix(Mode="Rec.601->Rec.709",clamp=false)

Looking at your script, it seems like you can't exactly decide what you're trying to do; sigma=0.6 won't be nearly enough to really remove all the grain. If you want to keep the grain as it is, use the latest version of grainoptimizer (which requires a heavily denoised input clip). Don't use the older versions; they've been found to artifact in some cases, which is probably something you want to avoid. If you want to remove the grain entirely, use a higher sigma value and higher block sizes. FFT3DGPU is good at this, IMO. Of course if you remove the grain entirely, the grainoptimizer is a bit pointless. Also, I've found the sharpen of FFT3D is quite good; you could try it instead of SeeSaw. Raising your BT parameter values might help also.

If you want maximum denoising effectiveness, use MVTools to do motion-compensated FFT3DFilter. The documentation has an example of this using RemoveGrain, but you can put anything in its place and it'll still work.

[tetsuo55]

Won't such strong denoising destroy details?

Maybe i should post a little sample clip??

do these make more sense?


FFT3DFilter(sigma=2, bw=48, bh=48, bt=5, ow=16, oh=16, dehalo=1, hr=1, sharpen=0.9)
grainoptimizer()
ColorMatrix(Mode="Rec.601->Rec.709",clamp=false, opt=0)

Or when using FFT3DGPU

FFT3DGPU(sigma=2, bw=48, bh=48, bt=5, precision=2, ow=16, oh=16, dehalo=1, hr=1, sharpen=0.9)
grainoptimizer()
ColorMatrix(Mode="Rec.601->Rec.709",clamp=false, opt=0)

[Morte66]

Won't such strong denoising destroy details?

A good denoiser will destroy a little real detail and a lot of grain/noise (which gives a false impression of detail). But yes, denoising will always remove some real detail with the noise and it's a judgement call. It's a matter of personal taste, and also your playback setup. Nobody can choose for you, you have to try a few methods and see what you like.

If you don't like noise, I'd denoise it then perhaps sharpen it (to get your false impression of detail by a different route). If you do like grain/noise, just use GrainOptimizer to help compression.

As for denoisers, if you're going to motion compensate fft3d then you may as well go the whole hog and use mvdegrain2 but with blksize=16 and overlap=8 (three times the speed of the default settings for a very small quality drop).

As for blocks, deblock_qed_mt2 (http://www.avisynth.org/mediawiki/wiki/Deblock_QED) is good for reasonable sources.

My usual DVD cleanup looks like:

# source

# deinterlace if needed

#TV (16..235) -> PC (0..255) levels. Much better blocking and banding in encode but your display must be adjusted for the different brightness/contrast to source DVD.
Tweak(bright=-16, cont=1.164, coring=false)

#deblock before any denoise/crop/resize to give deblocker a clean shot
DeBlock_QED_mt2( quant1=20, quant2=24, aOff1=2, bOff1=4, aOff2=4, bOff2=8 )

#crop black borders
crop(2,4,-6,-8) #or whatever

#resize up to next multiple of 16 for encoder efficiency and compatibility with GradFunkMirror() later
spline36resize ( 16*((width+15)/16), 16*((height+15)/16) )

#Denoise or grainoptimizer to taste
#grainoptimizer()
#FrFun7(1.1,2.5,1.5) #fractal spatial denoiser
#TTempSmooth(maxr=5) #pure temporal
#fft3dgpu(sigma=1,bh=32,bw=32,bt=4,precision=2,plane=4) #fast spatio-temporal, bw/bh=32 avoids occasional colour glitches
mvdegrain2p (source=last,blksize=16,overlap=8,sharp=2,thSAD=400,idx=1) #mocomped temporal, the daddy

#sharpen if you want (not my thing, ask somebody else)

#Tackle banding in flat areas
GradFunkMirror(1.2)


function MVDegrain2p(clip "source", int "blksize", int "overlap", int "sharp", int "thSAD", int "idx")
{
blksize=default(blksize,8) # blksize value (4, 8 or 16)
overlap=default(overlap,blksize/2) # overlap value (0 to half blksize)
sharp=default(sharp,1) #
thSAD=default(thSAD,400) # higher risks ghosting, lower risks "staggered denoising"
idx=default(idx,1) # use various idx for different sources in same script

backward_vec2 = source.MVAnalyse(isb = true, delta = 2, pel = 2, blksize=blksize, overlap=overlap, sharp=sharp, idx = idx)
backward_vec1 = source.MVAnalyse(isb = true, delta = 1, pel = 2, blksize=blksize, overlap=overlap, sharp=sharp, idx = idx)
forward_vec1 = source.MVAnalyse(isb = false, delta = 1, pel = 2, blksize=blksize, overlap=overlap, sharp=sharp, idx = idx)
forward_vec2 = source.MVAnalyse(isb = false, delta = 2, pel = 2, blksize=blksize, overlap=overlap, sharp=sharp, idx = idx)
source.MVDegrain2(backward_vec1,forward_vec1,backward_vec2,forward_vec2,thSAD=thSAD,idx=idx)
}


Function GradFunkMirror( clip c, float "strength" )
{
#counters edge effects for gradfun2db()
#gradfun2db() requires width mod8 and height mod2
strength = default( strength, 1.2 )
w = c.width()
h = c.height()
vflip = c.FlipVertical()
hflip = c.FlipHorizontal()

stackhorizontal( hflip.crop( w-16, 0, 16, h ).addborders( 0, 16, 0, 16 ),
\ stackvertical( vflip.crop( 0, h-16, w, 16 ), c, vflip.crop(0, 0, w, 16 ) ),
\ hflip.crop( 0, 0, 16, h ).addborders( 0, 16, 0, 16 ) )
gradfun2db( strength )
crop( 16, 16, -16, -16 )

Return last
}

Pick a few clips (use trim() statements in your script) with light and dark scenes, smooth areas, highly detailed areas, and fast movement. Then give the options a whirl and see what floats your boat.

[tetsuo55]

thanks for the advice so far guys

My videocard has died so i cannot try anything yet

However i did manage to upload a sample clip

The clip has dark and light parts, imho the grain is even worse than i thought

50mb sample:
http://www.sendspace.com/file/2sj4tm

Edit:

Looks like its interlaced too?

[Morte66]

That's progressive.

I'd be happy with what I already posted to process that, but you need to try it for yourself once your computer is back in action.

The only other thing I'd do is reduce it from 25 to 24fps and correct the audio pitch, but most people aren't as bothered about PAL pitch errors as me.

[tetsuo55]

Thanks

hmm pal pitch error??
Can you explain?

[Morte66]

Thanks

hmm pal pitch error??
Can you explain?

Film and a lot of TV (less in Europe) is shot at 24fps. PAL works at 25fps. One is usually converted to the other by just speeding it up, the video plays faster and the audio is higher pitched. Most people don't notice (e.g. you weren't aware of it), but some of us have a cow over it. So I use AviSynth to slow the video back to 24fps, drop the audio pitch to what it should be, and resample the audio for compatibility.

So my normal script starts like: video=DGDecode_mpeg2source("C:\meGUI\SG1\SG1.d2v") #video
audio=nicac3source("SG1 T01 2_0ch 224Kbps DELAY -128ms.ac3",2) #audio
audiodub(video,audio) # composite audio/video stream
delayaudio(-0.128) # adjust sync like DGIndex told us to in filename
assumefps(24,1,true) # redefine video as 24fps and audio as 46080Hz, slowing both to be correct
ssrc(48000,false) # resample audio to 48KHz for hardware compatibility

#rest of script does video stuff (denoise etc)
You can feed that AviSynth script to the audio and video inputs of meGUI et al.

Just try it a couple of times and see if you care.

[tetsuo55]

morte66,

Could you please post all the needed files for both those scripts?

Megui and DVd rebuilder are giving me errors about everything

I installed avisynth, megui, dvd rebuilder and masktools

[Morte66]

morte66,

Could you please post all the needed files for both those scripts?

Sorry, I can't because some of them have licenses that say they must be redistributed with source code and I don't keep that. But hopefully I can get you started...

A lot of it is internal or installed by meGUI, make sure you have AviSynth v2.57 (I don't think meGUI touches that).
For Deblock_qed_mt2 see the wiki link in my earlier post.
For FrFun7: http://soulhunter.chronocrossdev.com/data/frfun7_rev6.zip
For TTempSmooth, fft3dgpu, MVDegrain, gradfun2db and lots more see here: http://avisynth.org/warpenterprises/
GrainOptimizer you know.

If anything is still missing, Google.

[tetsuo55]

Deblock_qed_mt2 wiki link does not work, i cannot find the filter in google (they all point to the non working wiki)

I now have the following files

fft3dgpu_20060614.zip
frfun7_rev6.zip
gradfun2db_25_dll_20060315.zip
mvtools_25_dll_20060605.zip
ttempsmooth_25_dll_20051117.zip
tweakcolor_25_dll_20040412.zip
Avisynth_257.exe
megui-0.2.5.1007.exe
GrainOptimizer 2.01.zip

I found some newer versions though (not sure if this will break the script you posted?)
fft3dgpu0.8.2.7z
megui-0.2.6.1034.exe

That audio fix part does not require any extra plugins right?

Can i just cut and paste that to source like this:


# source
video=DGDecode_mpeg2source("C:\meGUI\SG1\SG1.d2v") #video
audio=nicac3source("SG1 T01 2_0ch 224Kbps DELAY -128ms.ac3",2) #audio
audiodub(video,audio) # composite audio/video stream
delayaudio(-0.128) # adjust sync like DGIndex told us to in filename
assumefps(24,1,true) # redefine video as 24fps and audio as 46080Hz, slowing both to be correct
ssrc(48000,false) # resample audio to 48KHz for hardware compatibility

# deinterlace if needed

#TV (16..235) -> PC (0..255) levels. Much better blocking and banding in encode but your display must be adjusted for the different brightness/contrast to source DVD.
Tweak(bright=-16, cont=1.164, coring=false)

#deblock before any denoise/crop/resize to give deblocker a clean shot
DeBlock_QED_mt2( quant1=20, quant2=24, aOff1=2, bOff1=4, aOff2=4, bOff2=8 )

#crop black borders
crop(2,4,-6,-8) #or whatever

#resize up to next multiple of 16 for encoder efficiency and compatibility with GradFunkMirror() later
spline36resize ( 16*((width+15)/16), 16*((height+15)/16) )

#Denoise or grainoptimizer to taste
#grainoptimizer()
#FrFun7(1.1,2.5,1.5) #fractal spatial denoiser
#TTempSmooth(maxr=5) #pure temporal
#fft3dgpu(sigma=1,bh=32,bw=32,bt=4,precision=2,plane=4) #fast spatio-temporal, bw/bh=32 avoids occasional colour glitches
mvdegrain2p (source=last,blksize=16,overlap=8,sharp=2,thSAD=400,idx=1) #mocomped temporal, the daddy

#sharpen if you want (not my thing, ask somebody else)

#Tackle banding in flat areas
GradFunkMirror(1.2)


function MVDegrain2p(clip "source", int "blksize", int "overlap", int "sharp", int "thSAD", int "idx")
{
blksize=default(blksize,8) # blksize value (4, 8 or 16)
overlap=default(overlap,blksize/2) # overlap value (0 to half blksize)
sharp=default(sharp,1) #
thSAD=default(thSAD,400) # higher risks ghosting, lower risks "staggered denoising"
idx=default(idx,1) # use various idx for different sources in same script

backward_vec2 = source.MVAnalyse(isb = true, delta = 2, pel = 2, blksize=blksize, overlap=overlap, sharp=sharp, idx = idx)
backward_vec1 = source.MVAnalyse(isb = true, delta = 1, pel = 2, blksize=blksize, overlap=overlap, sharp=sharp, idx = idx)
forward_vec1 = source.MVAnalyse(isb = false, delta = 1, pel = 2, blksize=blksize, overlap=overlap, sharp=sharp, idx = idx)
forward_vec2 = source.MVAnalyse(isb = false, delta = 2, pel = 2, blksize=blksize, overlap=overlap, sharp=sharp, idx = idx)
source.MVDegrain2(backward_vec1,forward_vec1,backward_vec2,forward_vec2,thSAD=thSAD,idx=idx)
}


Function GradFunkMirror( clip c, float "strength" )
{
#counters edge effects for gradfun2db()
#gradfun2db() requires width mod8 and height mod2
strength = default( strength, 1.2 )
w = c.width()
h = c.height()
vflip = c.FlipVertical()
hflip = c.FlipHorizontal()

stackhorizontal( hflip.crop( w-16, 0, 16, h ).addborders( 0, 16, 0, 16 ),
\ stackvertical( vflip.crop( 0, h-16, w, 16 ), c, vflip.crop(0, 0, w, 16 ) ),
\ hflip.crop( 0, 0, 16, h ).addborders( 0, 16, 0, 16 ) )
gradfun2db( strength )
crop( 16, 16, -16, -16 )

Return last
}

[Morte66]

Deblock_qed_mt2 wiki link does not work, i cannot find the filter in google (they all point to the non working wiki)

Well, it seems OK for me to upload so put this file (http://www.joel-benford.co.uk/temp/deblock_qed_mt2.avsi) in your avisynth plugins directory. I won't swear that's the latest version, but it works for me. You might find it complains about wanting even more filters, but you can chase them down.

[When you start up with avisynth, it's a good idea to comment out the bits that are tricky to install then add them back in one line at a time.]

I found some newer versions though (not sure if this will break the script you posted?)
fft3dgpu0.8.2.7z
megui-0.2.6.1034.exe

I think they should be fine.

That audio fix part does not require any extra plugins right?

It's all internal AviSynth functions except nicac3source, which I'm pretty sure comes with meGUI.

Can i just cut and paste that to source like this:

Yes. But note:
- You might get a different audio delay for the whole file from the cut section, so you need to check the filename and adjust the script.
- You can put the two function definitions at the end in files called "GradFunkMirror.avsi" and "MVDegrain2p.avsi" in your avisynth plugins directory so you won't have to paste them into every script, you can just call them.
- {edit} And make sure you pick the right audio track, I went for French!

[I ran the encode from that script, and there's a lot of banding broken into crawling miniblocks. GradFunkMirror(1.2) on the encode side will help spread them evenly, and if you use x264 to encode then some adaptive quantization (e.g. --aq-strength 0.4 --aq-sensitivity 15) will stop the encoder clumping them into nasty macroblocks. Then if you use the ffdshow video decoder, you can run GradFunkMirror again in its AviSynth section to counter banding caused by encoder quantisation. I find it's better to use a small dose of GradFunkMirror before and after a lossy encode than a large dose at either stage. But worry about this once you have AviSynth working.]

[RaynQuist]

FrFun7 gives the whole image a kind of "tint" like a chroma offset or something. I use FrFun3b (FrFun3d crashes for me with SetMTMode)

[tetsuo55]

So if i understand it correctly, the script causes "a lot of banding broken into crawling miniblocks"

My goals are 2fold:
Firstly i want to create a DVD5 of each DVD9 disk.
Secondly i want to make x264 encodes with subs, ac3 audio(and in later seasons commentary tracks) for my mediacenter

I was hoping to have a cleanup that would not require any scripts afterward.

[Morte66]

So if i understand it correctly, the script causes "a lot of banding broken into crawling miniblocks"

Well it reveals them, they're in the DVD MPEG2 under the noise.

You might prefer a lighter denoiser (e.g. fft3dgpu with sigma=1) to leave some noise to cover them. You might prefer GrainOptimizer to keep more noise, but store it efficiently. Or you might decide it's fine as it stands and just go with it. This is why you have to try it and see for yourself.

Secondly i want to make x264 encodes with subs, ac3 audio(and in later seasons commentary tracks) for my mediacenter

No problem, except if decide you'd prefer 24fps then that means reencoding the audio, and an encode to AAC/OGG would sound better than AC3. The chances are you won't care about that, most people don't but check it out.

I was hoping to have a cleanup that would not require any scripts afterward.

It doesn't require them, but it will get the very best results that way. Any denoiser you use is going to reveal banding from the DVDs and and lossy encode you make is liable to create some (though x264 with adaptive quantisation resists it very well indeed). You can either hide it with grain/noise like the original DVDs do or use a debander on playback. There is a DeBand function in ffdshow which works pretty well but it doesn't have full strength at frame edges. The GradFunkMirror() script fixes that, but it's more "extra points" than vital. Try all the options, see what you think.

[Zanejin]

You can either hide it with grain/noise like the original DVDs do or use a debander on playback.

I find that using ffdshow's DeBand filter helps significantly, but I'm quite curious about hiding the banding with grain. How would, say, using gradfun2db effectively, be accomplished with an encoder like x264, which really just seems to remove the added grain or simply artifact at bit-rates that correspond to CRF 18?

[Morte66]

I find that using ffdshow's DeBand filter helps significantly, but I'm quite curious about hiding the banding with grain. How would, say, using gradfun2db effectively, be accomplished with an encoder like x264, which really just seems to remove the added grain or simply artifact at bit-rates that correspond to CRF 18?

A lot of the work gradfun2db does pre-encode will survive x264 if you use something like crf 18 deblock 0,0 and adaptive quantisation, but not all. x264 doesn't throw away everything gradfun2db() does. IMO it's good to use it before encoding because it greatly reduces the odds of x264's banding (and blocking at band edges) reinforcing the original MPEG2 banding (and blocking at band edges). When those two get together, they can push things over the edge.

Once you decode your x264, there will be some banding (and blocking at band edges). You can reduce it with appropriate encoder settings but it will be there. Ffdshow's Deband will sort it out pretty well, provided you gave it something it can work with. Or you can use noise to hide it, or a bit of DeBand and a bit of noise.

As for noise... You've got two choices with x264: (a) don't denoise the original so much and adjust the x264 deadzone settings to make it retain more noise (which pushes the bitrate up for a given quality), or do a denoised encode and use the noise generator in ffdshow on playback to create "fake noise".

[tetsuo55]

Okay i must be doing something wrong

i formatted the pc with the new videocard, and its sortof working

Anyway:

I installed avisynth
I installed megui
I installed all the plugins to avisynth/plugins

Now when i try to load the avs script like you posted the first time (without the audio changes), every function gives me a "invalid argument" error, rendereing every part of the script useless

What am i doing wrong here?

[foxyshadis]

You must have made an error when you were editing the script. If you took out the AudioDub line, you have to replace it with just 'video', nothing else on the line.

[tetsuo55]

Well as i said i did not use that part.

Eventhough megui gives errors for every line of the script, dvd-rebuilder worked fine!

However the result with dvd-rebuilder was not so good, the image was all washed out, way too bright, and almost all detail was lost into the blurryness. I have to admit that 80% of the grain was gone but the reslting image was too bad to watch.

I'm going to play around with the settings a bit and see what i can make out of it

[Adub]

I think your use of colormatrix is wrong, at least if you want to be encoding to x264. If you want to encode to DVD, then it is fine, but I am pretty sure that AVC likes rec.601, not rec.709. That may help in fixing your colors.

[tetsuo55]

Thanks for your reply,

i used Morte66's script, and i encoded to dvd because megui doesnt work

So thats not the issue here