How to make good old QUALITY 700mb DVDRips?

[Damiano]

Ok guys, I've done a little more research and concluded that sceners definitely used the following individual tools:

- DVD Decrypter (if necessary)
- DVD2AVI / DGIndex
- NanDub / VirtualDubMod + AviSynth
- Xvid codec
- LAME MP3 encoder

I've also learned that VDM presumably as of version 1.5.x does support VBR (ABR) streams if you tick "By default try processing MP3 VBR streams like Nandub" in Preferences, which is hardly mentioned anywhere. Until then they would use NanDub strictly for muxing video and audio streams, which is why you can see the following in MediaInfo output:

Writing application : Nandub v1.0rc2
Writing library : Nandub build

Still, you can't produce ABR streams directly in VDM (the "correct" ones I mean), you have to extract the original audio track and convert it to MP3 using LAME typically with "-q 2 --abr 128" parameter and add it to the stream list before running 1st and 2nd passes. 1st to calculate the output file size with regard to audio size and 2nd for muxing while encoding video stream. Also, the reason why most sceners went for ABR as opposed to CBR and "vanilla" VBR is because it's the "smartest" and most compact mode, albeit a bit compromising in terms of compatibility or so they say. On the bright side, it will save up some bitrate for the video.

The rest is all about AviSynth filters. In fact, pretty much ALL modifications apart from pure re-encoding such as cropping, resizing, deinterlacing etc were done via AviSynth, never in VDM itself, because the latter doesn't even know how to address AR correctly e. g. if the source is 16:9 it will produce 4:3 output. Incidentally, that's also how Gordian Knot and AutoGK operate, though they won't let you use your own filters unless you pre-edit .avs files or produce audio streams with custom parameters beforehand. That's how I know sceners never used any of that.

Of course, it would've taken me much longer to discover all that if not for your kind input, guys. I also came across this ancient thread, which, seriously, I should've seen a long time ago. However, it's still not as complete as it looks and it doesn't mention some of the delicate bits I've shared with you just now.

Ok, so we have finally debunked it... not. The million dollar question remains and it's AviSynth filters. Which ones to use? Unfortunately, there is no way of knowing which combinations of those were applied by sceners as it's obviously not the kind of information you can retrieve from files. My major concern are 3 things:

- Deinterlace
- Resize
- Smooth/Deblock/Denoise/Deband

With regards to the first thing I'm completely clueless as I have little understanding of what interlaced content is other than a bunch of stripes appearing on motion with deinterlacing off in your player. But to be brutally honest, I don't really care much for what it is, I just need a good remedy for it. I know some filters just work like KernelDeInt() or TomsMoComp(), but I also want to know in what way do they differ and which one would make an optimal choice for most occasions. I've tried both and I can't seem to notice any difference, but I know it's there however subtle. I would like some input on this, please.

Resizing seems easier as at least some of those methods seem to differ dramatically. I came to know that Lanczos & Lanczos4 are not recommended for lower bitrates as their sharpness would in many cases lead to "ringing". Incidentally, that's one of the reasons why I dislike AutoGK, it maxes out on this thing and the output is always terrible. I have no idea which filters it uses, but probably something even sharper than Lanczsos / Lanczos4. I will have to see inside .avs it creates later. Anyway, after a few experiments I ended up being in favor of Bilinear and Blackman methods. First one seems like something most sceners had applied as it also does a very good job at smoothing the picture at the cost of little detail but with little to no ringing at all. It's great, though it's hardly ever praised. Blackman is something in-between Bilinear and Lanczos, it smoothens out the picture while keeping it sharp more or less. I've also tried Spline16/36/64, but failed to notice any substantial difference to what Lanczos does. So, I'm pretty convinced sceners mostly used either Bilinear or Blackman depending on the type of source, duration and target filesize. None of this is mentioned ANYWHERE. Still, if you have any advice regarding this particular aspect I'm all eyes.

Now, probably the toughest one as there is a myriad of different methods, filters and miscellaneous mods. They are also used in combinations which typically involves very delicate tweaking, depths, thresholds etc. This is way over my head. My other issue is that I can't always tell what's wrong with the picture, I can only see that something is off. Granted, I know what ringing and banding is, but what is "noise"? And how would "deblocking" help? Are these synonymous or not? I really need to know more about this, else I may have to dispense with it and stick with resizing filters alone, which, for all I know, will not produce 100% desired result. My rips are pretty good now, much better than they were years ago when I used AutoGK. But when I compare them to scene ones I can tell something is still missing. Something very delicate and it's definitely a combination of things. I could keep on experimenting I guess, but I'd rather I didn't spend too much time on this.

[zizouillette]

I have also discovered some new stuff here. First off, there is this much more profound tool than Gordian Knot, which still receives updates: MeGUI. It seems pretty complex and also better in a handful of ways. Anyone got experience working with it?

No way you can produce a better divx quality backup of a DVD Video with another tool than Gordian Knot
Firstly because it is not the software which determines the quality of the backup but the codec used, here the Xvid or the DivX.
Gordian Knot lets you access the codec settings which many software programs do not allow.
Do not bother too much the brain with the different settings of the codec because it is contradictory with a backup of a video weighing 4 or 8 GB in less than 1 GB
Make a backup of 1400 MB, we always stay in nostalgia because it corresponds to the size of 2 CD
The only software that I prefer to Gordian Knot is Xvid4PSP because the backup is done in one click with access to the codec settings and the possibility of applying filters ....

[Damiano]

No way you can produce a better divx quality backup of a DVD Video with another tool than Gordian Knot

If you had read my previous post you would know that the quality doesn't always depend on codec settings, there's quite a bit of AviSynth scripting involved, which alone can do plenty of magic. The truth is, Gordian Knot while being more flexible and less intrusive than AutoGK still cannot produce files as good as scene ones. There is no reason it should. And MeGUI at least offers more user-friendly control over these things.

Speaking of AutoGK, I decided to give it another go just to generate the .avs file and here is what I discovered:

LoadPlugin("C:\PROGRA~2\AutoGK\DGMPGDec\DGDecode.dll")
LoadPlugin("C:\PROGRA~2\AutoGK\filters\autocrop.dl l")
LoadPlugin("C:\PROGRA~2\AutoGK\filters\RemoveGrain SSE3.dll")

movie = mpeg2source("F:\...")
cropclip = autocrop(movie,mode=0,wmultof=4,hmultof=4,samples= 10,aspect=0,threshold=34,samplestartframe=0,leftad d=0,rightadd=0,topadd=0,bottomadd=0)
fixed_aspect = 1.18518518518519
c_width = width(cropclip)
c_height = round(height(cropclip) / fixed_aspect)
input_par = float(c_width)/float(c_height)
input_par = (input_par > 1.4) || (input_par < 1.25) ? input_par : (4.0/3.0)
out_width = 608
out_height = round(float(out_width) / input_par)
hmod = out_height - (floor(out_height / 16 ) * 16)
out_height = (hmod > 4) ? (out_height + (16 - hmod)) : (out_height - hmod)
new_aspect = (float(out_width) / float(out_height)) / fixed_aspect
autocrop(movie,mode=0,wmultof=4,hmultof=4,samples= 10,aspect=new_aspect,threshold=34,samplestartframe =0,leftadd=0,rightadd=0,topadd=0,bottomadd=0)
LanczosResize(out_width,out_height)
RemoveGrain(mode=2)

As you can see, it does use some plugins, which though are nothing special. Also, I was wrong about "sharper" filter, it's just good old Lanczos after all. As for smooth/deblock/deband etc it's only RemoveGrain, which probably doesn't even do a lot. This would explain why the output is so mediocre, other than computing resolution/aspect and performing autocropping there is nothing of great value here.

[zizouillette]

What determines the quality of a compressed video is Video source, Video resolution, Video bitrate, Video framerate, Video codec used.......This is valid for all video compressed.
With Avisynth filters you can increase the quality of some videos backups with filters like denoise, degrain, deblock.....But the settings that will be valid for a video will not be valid for another....
Do you really think sceners uses filters when they backup a DVD ?

@Damiano "If you had read my previous post you would know that the quality doesn't always depend on codec settings, there's quite a bit of AviSynth scripting involved,"

Thanks so much ......I was a member of an old French team who ripped hundreds of DVDs and made them available on DC ++ with our own hubs.
All used Nandub with the DivX3.11 codec, VirtuaDub with the XviD codec from Koepi and the DivX5.02 pro and is famous spyware Gator
After we used VirtualDubMod, FlasKMPEG, FairUsePortable, GordianKnot, RIAM, DVDx, XMPEG, Vidomi, Catalencoder, Neodivx, drdivx2, SimpleDivX, Divx Video Duplicator......
We have used the filters very little for the only reason that we are never sure of the final result. Some used the cropping function, others no, because some players displayed the video in wrong aspect ratio, some rip in 700 MB, others in 1400 others in 712 because video could be burned on a CD with the overburning function of Nero, some encoded the credits at 20% of the bitrate used for the rest of the video....Thanks again

[davexnet]

Wow, DC++, I was a user myself almost 20 years ago.
I think almost all mainstream commercial DVD's were "good enough", they didn't really need
filtering, except typically IVTC/cropping.

[Damiano]

What determines the quality of a compressed video is Video source, Video resolution, Video bitrate, Video framerate, Video codec used.......This is valid for all video compressed.

And duration.

With Avisynth filters you can increase the quality of some videos backups with filters like denoise, degrain, deblock.....But the settings that will be valid for a video will not be valid for another....

Well yes, obviously, which is why one must have options.

Do you really think sceners uses filters when they backup a DVD ?

Honestly, I can't be 100% certain but I thought some rips look too good to be just direct re-encodings from a DVD. But then a lot of them were made before 2005, which means filtering wasn't a big thing yet, so maybe you have a point, after all.

Thanks so much ......I was a member of an old French team who ripped hundreds of DVDs and made them available on DC ++ with our own hubs.

Sorry, I didn't mean no disrespect and I also wasn't aware of your experience.

We have used the filters very little for the only reason that we are never sure of the final result.

Thank you, that's very good to know.

Wow, DC++, I was a user myself almost 20 years ago.

You will probably laugh but I started using it just 2 years ago. It's not like I hadn't known about it, just never got around to it until then. Besides, I've been using emule since 2006 and had little need for another p2p network

[Damiano]

Ok guys, I think I've finally got the hang of it more or less. It appears that most of the time people wouldn't even change the default AS@L5 profile settings, which frankly are just ok as are on most occasions. I've fiddled with XviD settings a lot but I failed to notice any difference in the output, so I decided to keep the default ones. Besides, I still don't understand what does what exactly, like those b-frame settings and whatnot, I'm afraid so far it's beyond me.

Anyway, I'm still puzzled about resize filters. I know what they do exactly as I can clearly see the resulting effect, but I'm not sure about the actual convention for different types of media. What would be a good setup for a cartoon show like Johnny Bravo? I've already encoded season 1 and 2 using lanczos method, but the output seems a little too sharp. Yet I can't say the picture is bad it just looks unconventional. In fact, it's so sharp it reminds me of a wet sponge, especially season 2, which is a lot more crisp than 1. Is bilinear or bicubic method a good substitute here? Both seem to make the picture softer, but I'm trying to pick the best option here. Any advice? Also, does "Cartoon Mode" in Zone Options do anything at all? I failed the detect any difference.

[zizouillette]

When you do a DVD backup: never but never resize the video. No reason unless you want to watch your video on a watch.

Bilinear sampling is best used when reducing an image. The color of each pixel in the resized image depends on where the pixel would fall in the source
image.The color is interpolated from the four surrounding pixels based upon the theoretical position of the new pixel.
Because bilinear sampling considers only 2x2 pixels in the source and interpolates linearly it produces asoftened image on resizing and can cause pixelation in enlarged images.

Bicubic sampling is more complex than bilinear sampling, calculating the color of each pixel in the resized image from 4x4 pixels in the source. instead of simply interpolating linearly between
pixels bicubic sampling weights the contribution of each pixel in the source using a spline function based upon the offset of each pixel from the pixel being calculated.
Changing the spline parameters allows the sharpness of the resized image to be controlled. Three presets - soft, normal and sharp—are available to choose from.
Bicubic sampling produces better quality over bilinear sampling when enlarging an image but because more samples are taken and calculations are more complex for the bicubic method it also takes longer to compute.

Lanczos doesn't exist when i do rips

Cartoon mode: when you use 1 pass quantizer reduces the final video size by 3 or 4%, using 2 pass no differences, don't use this option

If you only have three or four DVD copy to make you can test all the options of the xvid encoder and also try some avisynth filters
But keep in mind that a scener who makes 5 or 6 movies per day will remain with the basic options, in fact ripping each DVD is long, very long .... encoding them is very long also especially in 2 passes ... checking the divx is also very long: is the soundtrack offset? good soundtrack selected? film encoded in full with credits and everything?
And I don't even talk about times when virtualdub crashes or displays an error message

The basic 2-pass xvid codec options are good for a DVD rip because DVD video is the best source to do a 720 divx

[Damiano]

When you do a DVD backup: never but never resize the video. No reason unless you want to watch your video on a watch.

As the topic indicates, I'm making 1CD rips, meaning I have to downscale the video at least to maintain the correct aspect upon cropping.

Bilinear sampling is best used when reducing an image.

I noticed that provided you have a high enough bitrate (1000kbps+) you can use lanczos/lanczos4/blackman and have very few artifacts. Bilinear in some cases makes the picture softer and reduces ringing, and in other it actually makes things worse. I'm not sure why. Which one would you use though?

Bicubic sampling produces better quality over bilinear sampling when enlarging an image but because more samples are taken and calculations are more complex for the bicubic method it also takes longer to compute.

That's interesting because some of these 1-click converters like AVI.net or AVI Re-Comp attempt to use Bicubic resize by default for downsizing. Did their authors not know it's incorrect? I'm actually surprised that these people were resourceful enough to create these apps to start with but used seeminly arbitrary and downright wrong settings for both xvid and avisynth. It's partly the reason behind AutoGK's low quality output, its built-in preset for xvid is totally out of whack (it uses MPEG-2 matrix for any file size to begin with).

Lanczos doesn't exist when i do rips

Then which resize method would you recommend for a 1CD (700mb) rip? Bilinear?

Cartoon mode: when you use 1 pass quantizer reduces the final video size by 3 or 4%, using 2 pass no differences, don't use this option

Ok, thanks. What about grayscale? What if I want to re-encode some b&w movies, is using it to any advantage at all? And what does 'chroma optimization' do?

But keep in mind that a scener who makes 5 or 6 movies per day will remain with the basic options

I figured that and like I said earlier I couldn't see any difference between default AS@L5 profile settings and tweaked (VHQ modes, Adaptive Quantization, i-frames etc), so I doubt I will ever touch any of that stuff again unless I know/see exactly what it does. I really hate feeling dubious and insecure so I trust xvid author provided the optimal default configuration, which should be good enough as it is.

And I don't even talk about times when virtualdub crashes or displays an error message

Yes, it tends to crash, but thankfully never while at a job (phew!).

Ok, after playing a bit more with Gordian Knot I came to realize that it really is the best tool for xvid. Not too comfy, but it does just what I need. What I particularly admire about it is the aspect error calculations, which is very useful when doing manual cropping and picking the right resolution and pretty accurate overhead calc, which is something a lot of apps miss. It's really cool. I'm still puzzled about the compressibility check though. Does it help the process at all or is it there to indicate critical compression ratio so you could do something about it like downsize the video a bit etc?

The big question still remains, though: which filters to use and how? What are the optimal and more or less universal parameters? Here we face a tripple challenge:

1. Too many filters plenty of which do either the same thing or slightly different
2. Too many possible combinations thereof
3. Seemingly complex sets of parameters where typically the more effective the filter the more complex (something like FF3D for instance)

I'm trying to figure out which filters and combinations were most popular in those days. I don't really feel like spending the rest of my life trying out each one of them individually, then trying out different combos etc. I want something relatively simple, effective and 9 times out of 10 applicable. At this point I have concluded that I can use at least the following 2 at all times:

RemoveGrain(mode=2)
FluxSmoothST(7,7)

They seem to do the job of both deringing and debanding quite well, I can definitely see the effect. But still, the resulting quality is just good and I know it can be "perfect". I'm afraid I'm pretty much stuck here on my own.

The less urgent question but still actual is which deinterlace method is the best or the least impactful to the quality? The 3 most popular methods are KernelDeInt, Field and TomMoComp. Even though they are documented I still don't understand what they do or how they are different as the result is always identical?

[Damiano]

Ok, guys, I have another question. Is it ok to crop video using resizers instead of crop() function? Someone here said it will 'blur chroma channels', whatever that means. I've used both crop() and lanczos internal cropping feature and couldn't see any difference. What I want to know is, is it safe to do it this way or not?

[azmoth]

Why bother fiddling with GordianKnot when you can do everything(cropping, deinterlacing etc) with AutoGK using 1 pass quality based and AGKPal to run your preferred Avisynth script?

[Damiano]

Why bother fiddling with GordianKnot when you can do everything(cropping, deinterlacing etc) with AutoGK using 1 pass quality based and AGKPal to run your preferred Avisynth script?

Well, first off, I'm not really fiddling with GK, I'm working with VirtualDubMod and AviSynth directly. Secondly, I don't really know how to make AutoGK use custom scripts and xvid settings, for all I know you have no control over those things and it will always do things its own way. And if autocrop() function is all there is to it then it's not exactly what I need anyway. It's a bit inaccurate and the material I'm working with right now has to be cropped manually one way or another.

[Damiano]

OK, sorry. I see what you mean now. I've never seen this AGKPal thing before, but I've failed to run it alongside AutoGK so far. Anyway, albeit interesting I don't really need it at the moment as it doesn't answer my earlier question with regards to cropping. I will make it real simple now: is there ANY difference between, say, crop(16,6,688,564) and internal resize cropping e. g. LanczosResize(640,480, src_left=16, src_top=6, src_width=688, src_height=564) ? The only thing I'm aware of is that I can only use even numbers with the former and ANY numbers with the latter, though, at the expense of.... what exactly?

[orion44]

I opened up a dusty box of old DVD's and pulled out Red Dragon, NTSC & anamorphic. Ripped with MakeMKV, then authored to
VIDEO_TS folder with AVStoDVD. Created d2v with DGIndex.
Logged onto my old XP system that has xvid installed and opened the following script in the original Virtualdub

loadplugin("F:\Program Files\AviSynth 2.6\plugins\TIVTC.dll")
LoadPlugin("c:\Documents and Settings\Dave New\My Documents\Downloads\dgmpgdec158\DGDecode.dll")
mpeg2source("i:\test\red_dragon.d2v")
tfm().tdecimate()
spline36resize(852,480)
crop(0,56,-0,-60)

Used the bitrate calculator at http://www.cole2k.net/bitrate-calculator.html and got a bitrate of 921kbps for 15 minutes,
128 kbps audio, target size 115 MB. Did a 2-pass encode, finished file 112 MB

Set up the Xvid codec as recommended by the guide. The result seems pretty good to me.
Here's a short clip, direct stream from the output

PS it feels really weird to be playing around with xvid/avi again after all these years

Your DVDRip looks excellent. Can you give me the exact XviD settings that you used for this rip? Did you use the XviD version with or without VAQ patch?

[davexnet]

I don't remember, but I did mention I used the settings as recommended by the "scene" guide from all those years ago.
(Actually it was put together by somebody from the "thebox" private torrent tracker, they specialized in old UK TV shows)
https://www.mediafire.com/file/fm49m077w10i97m/encoding_guidev2.1.pdf/file

There's also jagabo's recommendations which are similar
https://forum.videohelp.com/threads/339944-XVID-conversions-too-small/page4?#post2120344