Anime Settings

[midiguy]

Hey there. I've been messing around with some anime DVDs, specifically my Cowboy Bebop Boxset. Just wondering if anyone has a good method for encoding anime. When I say "method" I mean:
- filters (smoothers, etc. and the order to put them in.)

- settings (quants, RC averaging, RC reaction, rate control down/up, pshychovisual, quarter pixel?, resolution perhaps?)

2 more things:

1) bad cropping situation:
top = 4
left = 4
right = 8
bottom = 0
aspect error.. and I hate cropping footage, especially this much footage, any suggestions?

2) I am aiming for 175 MB eps with vbr 128 kbps MP3 audio.

any help would be appreciated. thanks.

*EDIT* oh yeah, and what about IVTCing? I used decomb with the default IVTC settings, but I hear that you should mess with some other things for anime like gthresh and all that? is this because of some extra combing around moving mouths and the etc.? because I did not notice much extra combing, the default IVTC seemed to do well... oh yeah, and I am talking about DivX 5.02 here (well if that wasn't obvious enough...) suggestions = appreciated, thanks.

[Mac Sidewinder]

Doing a search using the words "anime" and "setting" resulted in 48 matches. Isn't there something there that can help you at all?

Mac

[DJ Bobo]

First of all: no offence to Mac Sidewinder :D

To the "problem":
1) I have here the Cowboy Bebop US DVDs, there is no need to crop top and bottom, just some pixels left and right, that's ot.
2) The image is clean, so no need for *ANY* filtering. I don't filter anyway, filtering destroys the image IMHO. For noisy DVDs BilinearResize does the job ;)
3) No special DivX5 settings for Cowboy Bebop or not Cowboy Bebop: universal settings:
*2-pass of course
*Max Quantizer on 8
*Bidirectional Encoding
*All other options off and no psycho stuff (all decoration IMHO)
4) cropping is generally needed whenever you see black, so don't get on your own nerves trying to get exactly 4:3, crop as much as needed and resize to a "legal" 4:3 format, whatever GKnot says or not.
5) For 175MB per episode, 448x336 is the resolution you're looking for.
6) Using Decomb you won't EVER see any combing, because Decomb cheats: it uses FieldDeinterlace within Telecide to clean what Telecide missed.
If you ask me, I use IVTC 2.2 with the following settings:

ivtc(44,11,95)

99% of your video will be clean and smooth and works on Cowboy Bebop without problems.
If you're tooo sensitive of combing, combine IVTC & Decomb that way:

ivtc(44,11,95)
FieldDeinterlace()

[RadicalEd]

Bebop is generally clean but there is a lot of raibowing around black lines and stuff, so you might want to use SmartSmoother IQ, unless it dosent bother you. Decomb's 'cheating' dosent work for the mouth problem because the interlacing is so little that it isnt detected (but its really annoying to watch).

[midiguy]

I have the Asian box set, not the US box set. These DVDs are pretty clean too, but I would like very mild filtering, what is reccomeded for something very mild (with settings please)?. and my cropping problem still remains because I have the Asian box set not the US, so hmmm. I dunno, I cropped the black borders, no more, no less, and resized to 512 x 384, I don't care what gknot thinks, I get the feeling that gknots aspect ratio calculations are inconsistent for some reason.

P.S. I got the Asian boxset for $54 CANADIAN which is like $35 US so no complaints here.

please comment on the cropping + the filtering.

Also, last time when I tried to make the max quant 8 I got very ugly results for some reason.. not sure, gonna try again.. also, what about b-frames? thanks.

[DJ Bobo]

Well, as said I wouldn't filter it at all, but if you insist, you may use:

TemporalSmoother(1)

About rainbows: keep them where you find them! If BANDAI wasn't curtious enough to make a proper transfer, why should I bother?! if you get too excited, you'll destroy the image completely.

About cropping: crop where you see black, I don't understand where your problem is?! and 4,8,4,0 isn't much at all, it's rather small, you havn't seen much HongKong DVDs, have you? on some of them you may go up to 20 for right and left (and it's ok for me as long as I see no black, as said don't bother too much about aspect ratio)

As for mouth interlacing: I havn't noticed any (I don't even use FieldDeinterlace), but you may go for this if that makes you happy:

ivtc(44,11,95)
FieldDeinterlace(full=true)

Beat me if you see any combing with it!

About b-frames: what do you think "bidirectional encoding" is for? ;)

About Max quantizer: I've been encoding with 8 for a loooong time now, no disavantadges at all. I think you used some rather low bitrate, so your encoding was done with quantizer 8 all the time or so, that's why it looked nasty.

[Jas113]

What program are u guys talkin about for encodin, Gknot? Vdub? Thanx for clarifying :)

[Acaila]

GKnot=VDub :)

[HeSh]

@DJ Bobo:
hmmmm...
i just tried that ivtc(44,11,95) and i must say that it doesn't work perfectly... your previous options (you have posted them in some other thread) (60,11,100) worked better. However the best results I had was with decomb filter. (Telecide & decimate).


I found that smoothing is quite helpfull with Hong Kong DVDs since they have a certain noise level (US releases are mostly remastered).

BTW
How do I change quants in Gordian Knot?
I would like to set all the encoding options and let him do the encoding for n episodes... I realy like how it processes jobs.

[DJ Bobo]

@ hesh
Decomb is not better, it's just cheating, it uses postprocessing to hide its errors.
Use

ivtc(44,11,95) (or what ever settings you thing work better)
FieldDeinterlace()

Or to see how Decomb really is:

Telecide(post=false)
decimate(cycle=5)

and you'll see what I mean ;)
IVTC works better IMHO as I tested this and found that Decomb has mainly problems with motion -> too many scenes have jerky motion, especially dark scenes and scenes with much transparency like clouds or so. For example, the Arc The Lad DVDs from ADV: Decomb fails on the whole line! motion is jerky almost all the time. But with the optimized IVTC settings, motion is perfect almost all the time.

All IVTC filters fail though in DVDs with bitchy telecining. Look into the evangelion thread for my newest discovery ;) :D

[midiguy]

what is the advantage of using IVTC 2.2 over decomb when in decomb you can turn the post processing off?

isn't the post=false command maent so decomb DOES NOT deinterlace anything? so what do you mean by it sneaks some psot processing in? thanks!

[manono]

Hi-

He just means that IVTC has no deinterlacer built in and Field Deinterlace is built into Decomb by default, so you have to turn it off to do an apples to apples comparison. But you still have to add some kind of deinterlacer to IVTC to catch the interlacing that slips by.

But I still think he's wrong though, and that Decomb does a better job overall. I'm with Hesh.

[midiguy]

hmm. so should I put in post=false or take it out? will decomb deinterlace some good progressive frames by addident or anytihng like that if I don't put that in? I couldn't find ANY interlaced frames when I used decomb to IVTC this bebop ep 1, and I don't think I ahve found any slip through in the past.. so is it really neccary to not put post=false in? thanks!

[HeSh]

You should take it out...
unless you are compleatly sure there will be no interlaced frames.

I tested both ivtc (this time with FieldDeinterlace())& decomb (with post=true & false) and found this out:

1. post=false... it does good job, but every n frames i get interlaced frame or 2.
2. post=true... same as before but the interlaced frame is gone. The frame is a little blury though. :(. Perhaps this can be fixed changing other options of Telecide() function... but I don't have time to experimet much.
3. ivtc removes this frame or (most likely) it does not decimate like decomb :confused: ??? before it was frame number 7033,but now it is 7032. oh... yes... the frame is correctly deinterlaced... :)

Here is an example : picture is as follows: post=false,post=true,ivtc.
hmmm... why is my attachment not seen? oh, well you'll have to trust me on that.

[DJ Bobo]

@ manono
I don't know how much you're testing, but it's no rarity that I reencode a given episode at least 2 times, when I got bad results.
As Decomb came out, I used it as there was/is some sort of hype about it. But as said, it's not that good as people think

Beginning with the DBZ DVDs, Decomb does its job very good but motion is jerky in many points of the episodes. IVTC without even using FieldDeinterlace works properly all the time and makes far less errors (nearly 0), and adapts instantantly or very fast to every pattern change.

Now to the Arc The Lad DVDs, Decomb is absolutely awful! a lot a lot of jerky motion. IVTC works here exactly like in DBZ.

Next to the Trigun episodes that aren't telecined in a standard way: IVTC usually chooses the right frames with less interlacing artefacts, Decomb fails on the whole line, it takes a lot a lot of the strong interlaced frames.

I think poeple are blinded through the postprocessing of Decomb, because without postprocessing it shows its real weak points. Motion remains an issue anyway.

I just wanna tell you that I transcoded nearly 80 Anime DVDs till now, and the number is increasing every week, and I want highest quality IVTC. If Decomb was better, I would have used it, but no, it can't even compete with IVTC 2.2

Just wanna explain things:
First parameter (here 44): it tells how interlaced a frame have to be, to be considered so. Choose 43 and IVTC won't recognize interlaced frames properly.
Second parameter (here 11): it tells how clean a frame have to be, to be considered as progressive.Choose 10 or 12 and IVTC won't reconstruct fields properly.
Third parameter (here 95): it tells when to change pattern, and depends on the first two parameters. Choose 90 and it won't recognise the pattern change as it should.

[manono]

Hi D J Bobo-

We're all friends here, right? We're all trying to learn and produce the best anime and movies possible, right? So please take this in the spirit in which it is intended. I've changed my methods many times in the past when I've learned of a better way, and I'm sure I'll do it again in the future. I'm not married to any particular methods. I'm not sure why you felt the need to establish your credentials. I've read your many knowledgeable, if sometimes strident, posts all over the forums, and admire and respect your abilities. But since you seem to feel it was important...

I don't know how much you're testing,

As much as you, I'm sure. And I can back that up with hard evidence of programs we use in our encoding that I've worked with in their early stages.

...I reencode a given episode at least 2 times

Same here-my record so far is 25 or so complete encodes of Jin-Roh (not because of IVTC, but because I was trying to get rid of the Edge Enhancement-pixelation without completely wrecking the picture in other ways). But the majority of my encodes will also be done multiple times before I'm satisfied.

I transcoded nearly 80 Anime DVDs till now

I counted mine up yesterday-86 anime DVDs and with the other movies, the total is about 250-more than most, fewer than many. And many of the Chinese movies also need some serious IVTC tweaking. This doesn't count the ones I've worked on multiple times.

Before going on I'd like to back up to some things you said earlier. Cowboy Bebop is something I know well, and commonly use for my testing.

there is no need to crop top and bottom, just some pixels left and right,

So it's OK to leave the black bar on top during the opening song? So it's OK to leave the remaining black bar at the top during some of the computer scenes in some of the episodes. It's OK to leave the remaining black bar on the left during the end song of some episodes? Maybe it is; it's a matter of personal preference, I guess. But for those that want to get rid of all black bars without cutting too deeply into the picture in other parts of the episode, there are other ways to accomplish it.

The image is clean, so no need for *ANY* filtering.

The image, while better than most anime, is anything but clean, but, again, it's a matter of personal preference whether or not to try and clean it up a bit. It comes with mosquito noise and rainbows, and the encoding adds more mosquito noise, block noise, motion artifacts, etc. The early episode vobs average 1.4 GB in the R1 CBs, but later they switch to around 900 MB. That lowers the average bit rate and the quality of the later episides.

...whatever GKnot says or not. and as said don't bother too much about aspect ratio)

In my opinion, GKnot is one of our greatest tools to help us get correct AR with very little difficulty. I don't much care what you do in your own encodes, but I do think you should be careful about the advice you give to others on this subject. Some people might actually pay attention to you.

Now-down to business. You are absolutely correct in saying that Decomb's greatest drawback is the jerkiness that sometimes (sometimes-not often) results from its use. Decomb, unfortunately, is unfinished. Donald Graft's next step before he was called away was to integrate Decimate into Telecide so that the 2 could "talk" to each other to better determine the duplicate frames to drop. Unfortunately, as it stands now, it sometimes chooses the wrong frame to drop which results in the jerkiness. As for reconstructing the frames, it is easily the equal of IVTC. Why? Because IVTC is entirely pattern based, and it doesn't always catch the pattern change in time or can't recover right away from bad edits. Decomb is basically field based, a much better system as it doesn't care about pattern changes, but you can turn on pattern based frame reconstruction, and determine the likelihood of pattern use overruling Field matching with the Guide and Gthresh settings. Anyone thinking that Decomb is no good based just on the Telecide-Decimate default settings is delusional. That's one of its beauties-that you can tweak to your heart's content to fit Decomb to the material.

And you are also correct in saying that Decomb can't handle Trigun (I've never worked with Arc The Lad or DBZ), but as we all know, Trigun is royally screwed up. I haven't had all that much experience with your IVTC2.2 settings. I know how to tune Decomb to fit my material, and it works very well most of the time, but neither Decomb, nor IVTC2.2 is perfect. When I find a passage that isn't working, I'll give IVTC2.2 a try, but it usually chokes as well. But I admit there may be material that IVTC2.2 does better on, just as there is material that Decomb will do better on. But I'll continue to experiment, and will try to use IVTC2.2 more often based on your recommendations. And I don't expect to sway your opinion, but I hope your experiments with it have been done with full knowledge of settings and how to use them. I think we'll just agree to disagree, but I thought the other side of the case should be presented also. Best wishes to you, D J Bobo, and keep up the good work.

[DJ Bobo]

@ manono
Just to let you know, I wasn't nervous as I wrote the last post. Sorry if you saw it as an offense or a challenge or something, my english is not that good after all ;)

Now to the discussion points:
First of all, I wanna say that I'm not a perfectionist or idealist. I'm the type of person that follows given quality standards without exagerating, that means I do whatever I can to get a high quality rip but I don't really care about perfection:
1) I don't care if some scenes aren't cropped correctly. I crop the minimum necessary. If some scenes rebel, it's not the end of the world.
2) I don't care about aspect ratio too much, I make it as close as possible and it's ok that way, even if I crop 10 pixel more or less, no one will notice anything anyway, so why bother getting it precisely?! I'm not making a high precision machine after all.
3) I think, starting from the moment where you filter, you're starting having high expectations from the image quality. I say: why bother because of some minor artifacts?! I won't see them anyway, as any normal person will watch from a distance of 2 meters at least. The only really worth-to-filter videos are VHS captures IMHO. For DVD I don't care (anymore)
4) the only thing I care for is IVTC and here again not too much, if one or 2 scenes get screwed up, it's ok, but if it is lasting almost all the time it gets on my nerves. I started experimenting because of the Arc The Lad DVDs. I tested Decomb with default settings and with the settings posted by Doom9, and both of them were decievent. So I switched (back) to IVTC and started playing with the settings around: I got higher quality & higher speed.
It is also very important for me to find universal settings, cause I rip almost every day 3 or 4 episodes. I was by 60,11,110 and then Awatef came with 44,11,95 and I don't think I can get the filter to work any better than that.
Of all things I tested, there is only one scene where Decomb works better: credits of a Saint Seiya Movie token from a HK DVD. The credits won't ivtc correctly with IVTC2.2 whatever I do! Decomb done it correctly. The movie itself and all other things I tested have better ivtc quality with IVTC2.2 though.

[manono]

Hi DJB-

No, I didn't take it as offensive-not at all. But I admit I did take it as a challenge. But that's good-what is life without challenges? I was happy to have the chance to share opinions and views with you. Good luck and see you in the forums.