Avisynth Help thread

[D404]

Originally Posted by TheFluff

wrong

Mind you FFMS2 doesn't like h.264 transport streams very much.

[Zergrinch]

I'm trying my hand at DVD ripping, and would like to get the best possible results that I can manage.



My source has a problem with noisy edges. The picture isn't really obvious, but zooming it to full on MPC really shows the problem if you watch the area encircled in blue.

Here is a 10MB snippet from the VOB, which I cut using MKVmerge. I did no re-encoding:
http://upload.jetsam.org/movies/cmb01.mkv

Currently, my meGUI-generated Avisynth script I have consists of:

# Set DAR in encoder to 22 : 15. The following line is for automatic signalling
global MeGUI_darx = 22
global MeGUI_dary = 15

DGDecode_mpeg2source("F:\!DVD Images\!Cooking Master Boy\Disc 1\01\VTS_01_1.d2v", info=3)
ColorMatrix(hints=true, interlaced=true, threads=0)
tfm(order=1).tdecimate(hybrid=1)

#deinterlace
crop( 8, 0, -8, 0)
#denoise

TextSub("F:\!DVD Images\!Cooking Master Boy\Disc 1\01\Episode 1 - Hardsub.ass")

What combination of Avisynth filters would you gentle folks recommend to optimize this?

[Scab]

TComb with appropriately tweaked settings (read the included documentation) should take care of most of it. edgecleaner() can probably do a decent final clean-up job after that.

[SHiN-gx]

I'm having trouble fixing this:

At every scene change(?), the last 2-3 frames of the previous scene or the next scene and are always blocky. The problem is the source itself. Is there an Avisynth filter can fix this?

Example:



EDIT: Also getting the same errors on 2 other DVD sources.

[max2k]

Try DeBlock or DeBlockQED. If you have a non rencoded DVD Source u could use the intern deblocking of MPEG2Source(CPU=/CPU2=).

Deblocking will smooth the Source. So if you use DeBlock or DeBlock QED and the Anime is realy worth you time, you could work with trim to filter the framechanges stronger.

[TiGR]

Or you could try applying the deblocking filter via mask, imagine MSCDetection.TemporalSoften. This would apply the deblocking both before the cut and after it, so it might be a better idea to duplicate the original mask and shift it by 1, 2, 3, 4 frames and then merge it into final mask with decreasing weight. I think it’s worth a try because trimming manually is only good for experienced YATTA freaks

(Not sure if MVTools’ scenechange detection is reliable enough with anime, though. It could just randomly deblock = blur things there and there.)

[SHiN-gx]

@max2k:

Setting the CPU made little difference and will try Deblock_QED() in a little why and...

you could work with trim to filter the framechanges stronger.

Please explain it to me in detail.

@TiGR:
TemporalSoften did the trick but it killed most of the details. Used lower settings but still to no avail. MSC Detection... searched that on Doom9 and it led me to RemoveDirt and led me to RemoveNoiseMC_HQ() and stopped using it because its requirements are conflicting with AnimeIVTC().

Tried DeDup() and on the default settings, it's not really helping.

[TiGR]

Originally Posted by SHiN-gx

TemporalSoften did the trick but it killed most of the details. Used lower settings but still to no avail. MSC Detection... searched that on Doom9 and it led me to RemoveDirt and led me to RemoveNoiseMC_HQ() and stopped using it because its requirements are conflicting with AnimeIVTC().

You want to deblock poorly compressed frames near scenecuts but avoid unnecessary loss of detail everywhere else. In other words, you want to merge two clips, the one with deblocking applied and the one without it.

MSCDetection is an MVTools2 function that outputs a mask: scenecut frames get white, all other frames are black. This is good, but you wanted to do the filtering on more than just one frame, which means you have to (temporally) smooth the mask first, hence the dumb TemporalSoften. Imagine the luma values of the mask: 0 0 0 255 0 0 0; after softening you will get something like: 20 100 160 255 160 100 20. This means that the deblocked original clip (ie. “deblocking itself”) will fade in and out at scenecut rather than blink in for one frame.

original = last
deblocked = original.deblock()
super = original.msuper()
vectors = super.manalyse()
scenecut_mask = MSCDetection(original, vectors).temporalsoften(3,255,255,255,2)

mt_merge(original,deblocked,scenecut_mask)

[max2k]

Originally Posted by SHiN-gx

@max2k:


Please explain it to me in detail.

Simple Version and this should be enough when you only need to change one filter. In your filter chain you insert something like this, instead the filter:

trim(0,278).
Deblock_QED(quant1=16, quant2=14)+trim(279,286).Deblock_QED(quant1=30, quant2=28)+trim(287,334).
Deblock_QED(quant1=16, quant2=14)+trim(335,342).Deblock_QED(quant1=30, quant2=28)+trim(343,0).
Deblock_QED(quant1=16, quant2=14)

The First trim defines the range: "trim(0,278)" form the start of your video (0) to frame 278. Then the dot and filter with the settings for this range: ".Deblock_QED(quant1=16, quant2=14)". "trim(279,286)" important the trims have to be aligned, the first segment ends with frame278 the second starts with frame 279.".Deblock_QED(quant1=30, quant2=28)" the same filter with other settings.

And so on ...

At the end "trim(343,0)" with filter and settings from frame 343 to the last frame(0)! of you video stream avisynth will using this settings.

The second way, would be making "functions" out of different filter chains, naming them and call them in the trims... Even more complicated...

If you don't use Yatta, there is still AvsP: http://avisynth.org/qwerpoi/ . A very nice Tool to edit and compare your scripts. Should be very helpful.

I now think I really suck at making sentence in English. But I think I understand it very good so I should better shut up and read only.
p.s:
At the moment I cannot call myself “YATTA-mad”, but hope to get my ass down this holiday season and do some manual IVTC and decimating to get a nice vfr encode out of some OAD that I planed to encode for a long time. Yes I am that ignorant that I really think I could teach my self YATTA, thers no one who would tutor me anyway.

[TiGR]

I’d suggest using a script like this:

q1weak=16
q1strong=30
q2weak=14
q2strong=28

trim(0,279-1).Deblock_QED(quant1=q1weak, quant2=q2weak)+\
trim(279,287-1).Deblock_QED(quant1=q1strong, quant2=q2strong)+\
trim(287,335-1).Deblock_QED(quant1=q1weak, quant2=q2weak)+\
trim(335,343-1).Deblock_QED(quant1=q1strong, quant2=q2strong)+\
trim(343,0).Deblock_QED(quant1=q1weak, quant2=q2weak)

Rather than repeating parameters over and over inline, I prefer to define them as variables – you can comfortably adjust them that way (no need to search-replace all the time). As for the trimming, I’d write down the frame numbers first and then paste them into the script without doing the math manually. The horrible trim-line is easier to edit that way. For the same reason, I’d split it into multiple lines (that’s what the slash does).

[[Kyuubi]Fansubs]

Hey,

Im having a bit of a problem.
I've extracted the VOB files from a DVDISO and then merged them into 1 VOB file with VOBMerge, to use it in MeGUI and encode it into an mp4 video.

But the result is that the picture gets "scanlines" over the entire picture when its in motion, but when its still or not moving very fast, picture is as its supposed to.

Is there anyway of fixing this with some Avisynth filter?
And does anyone know what causes this phenonomen?

Will attach these screenshots so youll get what I mean:

Spoiler for Screens:

Here is what it looks like in motion:


And here is the same scene seconds later when its not moving very fast or its stillimage:

[jfs]

The video is telecined and you need an inverse telecine (IVTC) filter to make it progressive.

Figure out the rest for yourself. (And if you don't want to spend some days learning about video theory, don't go into DVD ripping territory.)

[andy_blah]

Open the VOB with DGIndex first..., save the project, and from there create an AviSynth script loading the Yadif plugin (remember to load it with LoadCPlugin instead of LoadPlugin), then load also the DGIndex's dll file as prescribed at it's site and load it's project file with MPEG2Source("") then place .Yadif(mode=2, order=1) if the source is frame based and TFF (Top Field First, just make sure when you preview the video that you don't get jerky motion, if that happens switch to order=2 for BFF - Bottom Field First - or 0 to be automatically detected tho this isn't suggested for most cases), mode=1 is for field based (you have a double framerate than the vob's original). Basically you have to tinker with the settings until the source is properly deinterlaced. This is a clean and easy method that I prefer to use, and gets the job nicely done.

@ jfs: You always forget that there will be always a beginer needing for help, saying that doesn't help much.

[jfs]

Do not do what andy_blah suggests. Telecined video is not regular interlaced video, and the Yadif filter is for treating regular interlaced video, not telecined video.

[TheFluff]

Originally Posted by andy_blah

Open the VOB with DGIndex first..., save the project, and from there create an AviSynth script loading the Yadif plugin (remember to load it with LoadCPlugin instead of LoadPlugin), then load also the DGIndex's dll file as prescribed at it's site and load it's project file with MPEG2Source("") then place .Yadif(mode=2, order=1) if the source is frame based and TFF (Top Field First, just make sure when you preview the video that you don't get jerky motion, if that happens switch to order=2 for BFF - Bottom Field First - or 0 to be automatically detected tho this isn't suggested for most cases), mode=1 is for field based (you have a double framerate than the vob's original). Basically you have to tinker with the settings until the source is properly deinterlaced. This is a clean and easy method that I prefer to use, and gets the job nicely done.

@ jfs: You always forget that there will be always a beginer needing for help, saying that doesn't help much.

hi this is probably the worst solution possible, it is arguably even worse than doing nothing at all
needless to say, don't do this, do what jfs said instead

[salty]

andy, i think all of your encodes are failure if you have used yadif on telecine videos. this is what jfs and fluff were referring to when they said ivtc filters. check tivtc first because its the one that most people use when they do ivtc.

[[Kyuubi]Fansubs]

Originally Posted by jfs

The video is telecined and you need an inverse telecine (IVTC) filter to make it progressive.

Figure out the rest for yourself. (And if you don't want to spend some days learning about video theory, don't go into DVD ripping territory.)

Ah yeah, thanks alot.
Fixed the problem with Donald Graft's Decomb filter.
Slow process (took about 29 hours encode), but worth it.

Again, thanks!

[Zergrinch]

I'm running into some signs of overfiltering, and would like to get some feedback whether this is the case.

This scene is from the DVD:


I added a few basic de-interlace and cropping:
ColorMatrix(hints=true, interlaced=true, threads=0)
tfm(order=1).tdecimate(hybrid=1)
crop( 6, 0, -10, 0)

To make it look like this:


There is an annoying dot crawl that occurs, so as per advice from here, I added a few more filters:
tcomb(mode=2)
MCTemporalDenoise(settings="high",edgeclean=true,G PU=false)

Giving me this:


The script normally works very well. It's just this particular episode that's giving me a bit of problems, with the rainbowing and visible vertical blocks.

How best to proceed? Sample snippet from the DVD: Download

[TheFluff]

That's not dotcrawl, that's deinterlacing artifacts, and they're most likely caused by tfm failing to find a good match for that frame, so it's falling back on its postprocessing mode, which means deinterlacing the frame. Try with other tfm settings (use a mode that lets it try more matches) or drop it into yatta and find out why it's failing. If that fails too just use a better deinterlacer as postprocessor, like nnedi2.

You do however have a lot of rainbowing there, try bifrost if you want to get rid of it.


this is dotcrawl, by the way:


e: also, don't use colormatrix

[Desbreko]

TFM is deinterlacing because there's tons of dotcrawl which it sees as combing:



Nothing I tried managed to get rid of it, though.