People who have Evangelion US Release, please look here!

[DJ Bobo]

Motion is absolutely perfect, and there is no 30fps sequences in eva anyway. I tried the settings on episodes 12 & 13 of trigun and it works *like a charm*. Absolutely smooth motion! (absolutely no interlacing of course)
The only "problem" is slight line jitter, but that is normal if one uses Bob. If anyone knows how to get rid of this, please let me know.

I know that the english DBZ version sucks, I hate it if you ask me! I only care for the japanese version, above was just a test.
I did filter it slightly using TemporalSmoother(2). If I use SmartSmoother, I get 25% less size, in other terms 33% better quality by equal bitrate. But I can't allow myself to use this filter, since it takes too much time on my PIII@667MHz. But even so, it still needs a (very) high bitrate. (you can trust me on that matter, I have encoded *all* DBZ DVDs there is, I buyed them all :D)

[midiguy]

I am on a p3 600 mhz with 128 meg pc100 SDRAM. With my filter setup for dbz, this is what I got:

smart smoother (dia:5 thresh:30 weighdiff:3 mode:avereged pixels)
temporal smoother (2)
resize (640 x 480 precise bicucic -0.6)
subtitler (convert to SSA to add my own text (channel name, yes, I pirate, and the etc.))

and, I encode at a whooping 1 fps!!!!! wooo fucking hoo!
but, I don't care to wait the 17 hours for a nice looking 2 pass encode, or the 8-9 hours for a nice 1-pass quality based encode.

oh, by the way, spatial smoothing and temporal smoothing are 2 completely different things. so that is why I use both smart smoother and temporal smoother/ but I go way below the defaults for smart smoother otherwise I smooth too much. The way I have it set up above keeps ALL the details intact, I checked for 2 hours! and compared it all! all the details are there, extremely little unwanted blending. and, this makes the ep/movie about 40% more compressible, sometimes more sometimes less.

[DJ Bobo]

There was a time I used SmartSmoother, but not more because:
1) too long encoding time
2) gives a digital/plastic effect which I don't like
I'm thinking of not even using TemporalSmoother anymore, since it has a bad effect on pans with clouds and such stuff, don't know how to explain it well (clouds borders and the limits between color bands look excited/agitated somehow)
I think for DBZ BilinearResize does the whole job. It reduces noise without the effects I mentionned.
And 640x480 is a too high resolution anyway. With 3 episodes per CD, 512x384 is the right choice IMHO. (for 4 eps -> 448x336)
And I don't encode the subs in the video, since this cause quality loss (mosquito noise around the subs and higher bitrate consumption)

[manono]

Hi guys-

I hope you don't mind me butting into your private conversation. I think midiguy must be speaking of the SmartSmootherHiQ (even slower than the regular Smart Smoother). And since the groups require burning in the subs, then he hasn't much choice in the matter (although DJ Bobo is very right about separate subs being way better). I am surprised that they let you use DivX 5.02, since most groups still use DivX 3.11.

But the reason I'm writing is because it looks to me as if midiguy is using SShiQ and the Subtitler plugins of VDub, and therefore using the slower Full Processing. You do know that you can do those things in AviSynth and Fast Recompress without any color conversions, right? The SSHiQ for AviSynth is available here (http://cultact-server.novi.dk/kpo/avisynth/smooth_hiq_as.html), and the SSA subs can be put in with Textsub, which comes with VobSub. Who knows-by switching to AviSynth all the way you may boost the encoding speed to 1.5-2fps. :)

But if there's another reason why you're using VDub's filters and Full processing, then ignore me.

[Dreassica]

If i am not wrong it was stated that the avisynth port of SSMHiQ isn't faster as it's Vdub's version yet because it's not optimized yet!!

[midiguy]

Originally posted by manono
Hi guys-

I hope you don't mind me butting into your private conversation.

This was never a private conversation as it is on an open forum =)
and thanks for your advice about the avisynth. but, Dreassica is right that sshiq for avisynth is not optimized yet. so it actually results in slower speeds. also, I have a question:

Why would subtitles be HARDER to compress? wouldn't they be easier because they are just uniform colours and have extremely little detail? wouldn't actual video footage be WAY less compressible then subs? explain please! thank you! shalom for now!

[manono]

Hi Again-

I did some tests for you. I've done them in the past while beta testing SSHiQ, but no longer have the results, so I did them over again. This is the opening song of Cowboy Bebop plus 15 seconds of the episode that follows-1 minute 47 seconds altogether. Here's the .avs:

LoadPlugin("E:\DIVX STuff\GKnot\mpeg2dec2.dll")
LoadPlugin("E:\DIVX STuff\GKnot\Decomb.dll")
LoadPlugin("E:\DIVX STuff\GKnot\SmoothHiQ.dll")
mpeg2source("D:\Cowboy Bebop\Tests\CB-30fps.d2v")
crop(8,0,704,480)
Telecide(Guide=1,Gthresh=50,Chroma=True,Threshold=30)
Decimate(5)
BicubicResize(512,384,0,0.5)
SmoothHiQ(5,25,25,125,10)

I did it with XviD-Constant Quant 2 (I had something set up in DivX 5.02 that I didn't want to change). The AviSynth version of SSHiQ uses only Weighted Average (better than Ave. Pixel), while in VDub you have a choice of Wt. Ave. or Ave. Pixel. Other than that, the settings were the same (except Ave. Pixels turns off amount in VDub and the AviSynth version has both Luma and Chroma thresholds).

Fast Recompress-No SSHiQ.......1:52...........29,572 KB
Full Processing-No SSHiQ..........2:03...........29,168 KB
Fast Recompress-SSHiQ...........4:49...........26,688 KB
Full Processing-Ave. Pixel.........4:53...........27,192 KB
Full Processing-Wt. Ave...........5:34...........27,414 KB

You can draw your own conclusions. But if you're also using the Temporal Smoother in VDub, then switching to TemporalSoften2 in AviSynth will result in signifigant time savings. And if you're using the Subtitler in VDub, then you can switch to TextSub in AviSynth and I think you'll save some more time (not positive-never done any tests with that one). And as you all mentioned, after the optimizations are done for SSHiQ AviSynth, there will be even more reason to switch.

As for why it's harder to compress burnt in subs, I think DJ Bobo knows more about that than I. It's something to do with the encoder having a harder time compressing static subs when everything else is moving all around them. I don't think that anyone can argue against the fact of noise around the burnt in subs (although your filters will help that a bit). One thing I do know, though. I've done several 2 hour, 2 CD movies with constant Quant 2, and the file size of movies with burnt in subs is up to 50 MB larger than the same movie done the same way without the SSA subs. That's a significant difference (4-5%). I hope you didn't find this long post too boring.

[RadicalEd]

gah, sorry for being dead for a week but I'm back now. I'm trying the new thing right now but I wanned to say that I tried the original ivtc(44,11,95) thing again this time attempting to get the best motion (I was encoding at 320x240 so the interlacing didnt matter) and it seems to have worked perfectly as well, better than the 59.94 fps clip I had. I'll see if I can upload the clip in a lil bit, now I gotta rip another clip from eva and try bobo's method.

[ErMaC]

Well yes it turns out I am on this forum, I just don't visit here much.

Eva suffers from the dreaded "GAiNAX Syndrome". Basically, ever since Gunbuster way back when, GAiNAX has been using really really bad Telecine techniques. Not only that, but the people they actually have shooting the film are also very poor. Eva, Nadia, and to some extent KareKano all suffer from frame jitter (where they aren't keeping the various film cels aligned properly, i.e. the scene shifts around) as well as scene cut artifacts (those big white chunks at the top and bottom of the screen when there's a film splice) and very very bad Telecining.

As was described on the AVISynth documentation page, they do indeed use a method of Telecining where the frame change occurs in the middle of a field. This is why when IVTC'ing you will wind up with half of a frame on the top, in the middle there will be this weird blend of two frames, and then the bottom will be the next frame.

I would think that using AVISynth's Peculiar Blend option should help alleviate the problem, but I have never done this personally. I do happen to have the Eva DVDs on my hard drive right now, so I can go test it out myself if you guys want, but I would suggest experimenting with the PeculiarBlend option in order to fix the IVTC problems.

Bobbing and then Decimating is essentially losing you vertical resolution if I understand what you're doing correctly.

BTW, the Japanese Region 2 DVDs have the exact same problems in terms of film quality because the problem lies in the way Eva was shot on film and then Telecined to their video masters. It is not the fault of any other company.

Thankfully GAiNAX went all digital once Anime Ai no Awa Awa Hour (Ebichu) came around, so titles since then do not exibit this problem (FLCL, Mahoromatic, Abenobashi Mahou Shoutengai).

[ErMaC]

Well I've figured out how to get it to work, although I've run into a slight problem...

Either the top half or the bottom half of the frame will have its field order reversed. ^_^ And I can't figure out how to fix it.

Here's the script:

LoadPlugin("D:\Decomb.dll")
LoadPlugin("D:\MPEG2DEC.DLL")
MPEG2Source("E:\AMVs\Evangelion\EvaOp.d2v")
SwapFields()
SeparateFields()
PeculiarBlend(129)
Weave()
Telecide(chroma=true,post=false)
Decimate()

The other problem is that not every frame needs to be blended as such, and so some frames wind up getting worse instead of better, although this isn't very often nor very noticable.

But still, it's some progress. Does anyone have any ideas?

EDIT: The SwapFields() will just change which half of the picture is reversed. You can remove it, and then the other half of the screen will have the problem.

[DJ Bobo]

@ midiguy
I think, manono has answered your question about the subs. Encoding with subs, the files are getting bigger using constant quality mode ;) so subs are harder to encode. And don't try to speak theory too much, it's the same as credits, I always thought that credits take very little place because there is much black on them, but then it was the exact contrary ;)

About DBZ: I tried yesterday to encode with BilinearResize alone, it looked very good. So from now, I won't use any noise filters at all.


@ ErMaC
I'm not talking about scene change jitter, I'm talking about line jitter, that is typical for Bob. I'm aware of the image quality loss, but as far as I've seen, the latest method I posted is the only one that delivers smooth motion on the whole line. My first method with ivtc(44,11,95) delivers the best motion among automatic IVTC filters, but on some places not as smooth as the Bob/decimate method.
I think IVTC isn't the way to go anyway, since that was no telecining, but real framerate conversion, from 48 fields/s to 59,94 fields/s.
My Bob/decimate method consists in fishing out the cleanest 24 fields every second, and it works very nice.

[RadicalEd]

AAAH! Just like my dream O.O spooky. Now we just have to wait for Ddogg to come back... Ha, owell. Guess you got here through my post on the amv.org forums huh ermac. I just left home gosh darnit so I can't run tests with your method but it sounds good. Bobo, your method did give pretty perfect motion although the bob filter in avisynth does irreparable damage to the image quality if you ask me. So far the best yet (without having tried ermac's method) is still ivtc(44,11,95) and fielddeinterlace(). Actually I've been encoding at 320x240 so I havent been so worried about the extra interlacing, but ivtc(44,11,95) does look clearer than bob at 320x240 and gives the same perfect motion. I have a few clips that I'll upload tomorrow when I get home I suppose.

btw, just kidding about the it wasnt all good thing ^.^

[midiguy]

AnimeMusicVideos.org ownz. oh, and for anyone who cares, I have been making vids for a while now, but none were really good enough for release. but my current vid that I am working on is good enough I think, so I will be releasing it.

Anime: Streetfighter Alpha (no IVTC possible, damn HK DVD...)
Song: Crawling in the Dark (Hoobastank, I know, it is a bit too poppy for my tastes too, but whatever, works for the vid)

and yeah. when it is released I would like feedback, so please! thanks.

vid progress: 40% or something like that.

*EDIT* WHAT THE HELL!? for some reason I thought this was the Otaku Vengeance boards... forgot it was doom9. oopsies. oh well. probably because you were talking about AMV.org

[RadicalEd]

Hah, its okay. Its extremely humorous belonging to AMV.org's forums and these forums at the same time, because those boards are total chaos (in an extremely humorous way) and these forums are quite stiff and rule bent (not that thats a bad thing, there are also not many people on amv.org that will put up an intelligent discussion about video) but it would be great if people here posted stuff with topics like I like cheese or the whole dave and ana thing. Erm.. so anyway. I'm with you on the video front. I've sort of made a few videos (well... one and a half to be exact) but until now I haven't done anything releasable. I'm actually working on a new one right now and I pulled an all nighter 2 days ago starting work on another new one. Hmmmm.... well, just a sec. I'm puttin up an AMV section about my vids on my site right now so you can go check it out, just click the link in my sig. AAnd, finally. Right now I'm dling the avs wrapper so I can do auto ivtc and release the vid at full res (I really dont have the time or hdd space to save a bunch of clips at 640x480 and ivtc them in the mean time, plus the entire dvd is ripped already so might as well use it). Alright well, must go continue work. laters

just kidding.. the avs wrapper didnt quite work in premiere. Guess I'll have to go the old fashioned 320x240 huffyuv way _ _

just kidding again, theres an avisynth import filter for premiere? why didnt I know of this previously -- owell, thats a good thing.

[midiguy]

just kidding again, theres an avisynth import filter for premiere? why didnt I know of this previously -- owell, thats a good thing.

oh yes, that sure is a good thing. as for my sfa vid, I also didn't discover that AVS import filter until it was too late, and I don't have the space for 640 x 480 huffyuv video clips, so I am stuck with using 320 x 240 huffyuv clips.. oh well..

[RadicalEd]

I got sick of using the avs real quick, the IVTC and stuff was making editing like wading through a jell candle. I had around 8 gigs free on D (its rare, but I just cleaned recently) so I went with the 640x480 huffy. I'm about 2/3rds done with 3 gigs left so I should make it through if I'm careful.

[midiguy]

Originally posted by RadicalEd
I got sick of using the avs real quick, the IVTC and stuff was making editing like wading through a jell candle. I had around 8 gigs free on D (its rare, but I just cleaned recently) so I went with the 640x480 huffy. I'm about 2/3rds done with 3 gigs left so I should make it through if I'm careful.
good luck man!

[RadicalEd]

thanks :D actually I just finished but I can't put it online till monday for dumb reasons (prolly wont be able to get on till then either) so for all of u people, bye for the weekend :O

[Bionic-Badger]

Humm... I was messing around with that stupid ADV Eva DVD, going frame by frame through the raw data, and noticed some things about it.

Every six frames there is a perfect-condition progressive frame with no interlacing whatsoever. Between that the amount of interlaced material from each frame changes as the frames progress towards the next progressive frame. This appears to be classic "beat" patterning and consequently matches exactly a 24fps -> 30fps pattern. Ex:


0 1 2 3 4 5 6
| | | | | | | 30fps
| | | | | | 24fps
0 1 2 3 4 5


So I figured I could try using a frame selection method to have the frames on either "side" of the frame be used for raw field recombining and retain the progressive frame by simply making two copies of it so it would essentially be combined into itself.

So I made an AVISynth plugin to do that, but unfortunately, appending deinterlacing functions like weave, FieldDeinterlace, etc. aren't doing it... Any ideas?

I also found out that you can achieve *perfect* looking words and scrolling motion with the credits if this script is used (at least from the frame I was starting from):

doubleweave
SwapFields
SelectOdd

Unfortunately, the background image is messed up from the doubleweave.

-Bionic-Badger-

[ErMaC]

Cool that someone is still looking into this.

Interesting what you came up with. Unfortunately I'm now currently in Japan and can't do much research as I'm on a P2 366 and I don't have the Eva TV DVDs with me (just my own self-subbed movie DVD) but if you could post some examples of what your problem is I'd be happy to give feedback. I'm not quite sure what problem you're describing so if you could elaborate, maybe with some screenshots, that'd be great.

[Suzahara]

I also found out that you can achieve *perfect* looking words and scrolling motion with the credits if this script is used (at least from the frame I was starting from):

doubleweave
SwapFields
SelectOdd

Unfortunately, the background image is messed up from the doubleweave.

No, this doesn't seem to work at all. Reason being, you can't swap fields once your recover the progressive frames...all 4% of em in eva that I've found...but it messes up the picture too much by trying to swap the fields. It does work well on the outro though, nice smooth moving text, but it doesn't work on any other part and it doesn't seem to work on stationary text.

I've tried the other methods in here as well, they all leave a lot of ghosting, not a lot of interlacing, but ghosting where interlacing would be. Too bad not even manual ivtc in tmpg can handle eva 100%.

[Bionic-Badger]

I made an AVS plugin to try and extract as many progressive frames as possible, and it worked, at least, near the beginning, where the frames followed a horrible direct 5 frames -> 4 frames conversion. The problem is that it only works for a short while; later on doesn't work at all. It is like the pattern is changed on the fly according to whims or some pattern I don't see.

Even if all the frames followed the pattern that plugin was meant to handle, there is irrevocable damage to the frames, where there are fields that are blends of two different fields. In fact, the damaged field in the 24fps, four-frame blocks actually have data on two different fields in the 30fps sequence, about 75% of which is on one of them and 25% on the other. I could manage to pull two perfect frames out of each five-frame sequence, the other two had, each, one blended field, 75% intact. That would've cut the amount of blurring by 50% on top of whatever extra blending could've occurred. Unfortunately, the pattern changes.

-Bionic-Badger-

[Suzahara]

Yes, Eva doesn't have one pattern for the whole thing, I find that it changes every scene. The other problem is, some are impossible to convert to progressive and it seems the interlaced and progressive frame are blended together so you can't retrieve either...

One can only hope and find a way that gets eva 95% looking good, there's no way unless Gainax remasters the original that it will become a normal pattern.

[Bionic-Badger]

And even if they do, I'm not about to pay for another set of DVDs... I *DO* know that I'm not going to buy DVDs from ADV anymore...

-Bionic-Badger-

[Suzahara]

Well, I doubt another set will ever be put out. And I doubt you can find a set that doesn't have mastering problems as the original was this way too.

[ErMaC]

The original TV analog masters for Eva were like this - you're never going to get around the problem unless they go back and dig up the film which knowing GAiNAX may not even exist anymore. The Japanese releases have been this way since the LD releases, and the Japanese DVD exibit the same issues as the US ones (although the Japanese ones look better anyways). From what I've heard Eva was shot on 16mm film anyway - not even full 35.
You might see a remote chance of it happening - hell look at Macross. Animeigo went and dug out the original 16mm interpositives and cleaned 'em up.

[RadicalEd]

heh, I just realized that I'll be watching eva on a friend's progressive scan DVD player + HDTV sunday. The turnout should be interesting... if not horribly deformed.

[DJ Bobo]

@ Bionic-Badger
Not all ADV DVDs are like that, only the EVA DVDs are like that. I have the ARC THE LAD DVDs & the SAKURA DIARIES DVDs from ADV and they are telecined properly.

@ Ermac
You may be right about EVA being shot on 16mm, the original image is quite blurry.

@ RadicalEd
I think I know now why the motion on PC monitors is choppy. It's because the refresh rate is low.
When you produce a 60fps output from the EVA DVDs or any other telecined DVD (using Bob) and display it with 85Hz refresh rate, the motion can't be smooth.
The solution would be to lower the refresh rate to 60Hz or increase it to 120Hz.
My monitor can achieve 120Hz (which are more comfortable for the eyes ;)) when I lower the resolution to 640x480.
Try it and you'll see that the motion will become better.
BTW, a progressive scan DVD Player won't do any wonder ;)