anime with xvid

[triffid]

I have experimented with xvid on anime for quite a while over several of the more recent stable and unstable releases. Despite my excellent results on live action DVDs with the same codec, I can barely attain average results on anime (mostly edge artifacts and ringing, a bit of blocking). It appears xvid's settings are optimized for live action, and anime would require quite different settings. Anyway, does anyone have any settings that I could use as a starting point on anime source?

I used Nandub for a long time, and now have moved to Divx 5.02 and experimenting with 5.03.

Thanks.

-triffid

ps been a long time since I last posted, but never stopped lurking.

[Asmodian]

I use the default settings with B frames off, H263, and alt-curve compression at defaults (unless doing ~2 episodes a disk or something - then no curve compression). Most of the things I have noticed that need tweaking differently for anime relative to live action are avisynth filters, what filters are you using now? How big do you want your episodes? I usually shoot for 4 or fewer episodes per CD (@15 cents a disk why not?) so if you want it much smaller you might have different experiences.

I use decomb, one of the many good denoisers that are around if the source needs it (off a DVD it usually doesn't), convolution3d (for the compressibility increase), and neutral bicubic resize.

I don't get blocking at all (at least not more then the DVD had) and don't mind some light ringing as that is very hard to notice from 1 or more meters away from the monitor and is impossible to notice on the TV - but my TV is crap. Of course this is without post-processing. If you can't get a good encode try a slightly lower res, that helps a lot ;)

[triffid]

I use decomb, one of the denoisers, one of Donald Graft's new filters for sharpening anime (forget the name), and I experiment per episode on the others. I would use Conv3d if it had avisynth 2.5 support. That's not really the issue, as I use the same filters I would on my Nandub encodes, which came out quite a bit better than the ones I've been doing with Xvid.

I'll play with the guidelines you gave me, but I think you missed my point that, with the same input, I am unable to get satisfactory output with Xvid, even compared with old Nandub.

I appreciate your help

-triffid

[haibane]

Maybe you can try post out what setting you are using now so other people can have a better idea of what's wrong......
personally, I just keep my quants below 4.......
B-frame from 2 to 3 and 150 100.......
with qpel, LM, CM........

[Teegedeck]

Blocking? Strange, strange... What compression ratio do you aim at? I wouldn't go down more than 30% of the first pass (that's extreme, already, make that 40% ;)).

Originally posted by triffid
I use decomb, one of the denoisers, one of Donald Graft's new filters for sharpening anime (forget the name),

Sharpening may be the the reason you get very bad compressiblity here and resulting blocking. Try without it, perhaps it helps? It's a good idea to do compressiblity-tests (and keep the first pass to have a look at it) with different combinations of filters at least once if not for every encoding. BTW, do you use Marc FD's excellent MPEG2DEC3?

It could also be your curve-compression settings (linear seems the safest choice here, too).

[vlad59]

Originally posted by triffid
I would use Conv3d if it had avisynth 2.5 support.

-triffid

Convolution3D is available with avisynth 2.5

[Enchanter]

Originally posted by Teegedeck
[BMarc FD's excellent MPEG2DEC3?[/B]
Where can I find it?

Also, can someone tell me the actual differences between the different versions of MPEG2DEC.dll filters? They don't seem to offer much performance advantage over one another. Thanks.

[Lothar]

The problem I noticed more is the high compression used in dark areas of image: if it's only a small part of the scene there is no problem, but if the dark area takes the bigger part...

I'm using stable and unstable builds, whitout lumi, bframes or qpel and whit standard impos, and encoding a 22-minutes long anime episode.

[Teegedeck]

Originally posted by Enchanter
Where can I find it?

It's part of the Avisynth plugin package that goes with Avisynth 2.5.
Search the Avisynth forum for the 2.07 version.

Also, can someone tell me the actual differences between the different versions of MPEG2DEC.dll filters? They don't seem to offer much performance advantage over one another. Thanks.
Well, you could spare me some time if you tried to use the search button, but as I feel talkative today...;)
I got no idea what MPEG2dec2 can do.
Nic has written MPEGdecoder.dll which can open VOBs directly!
MPEG2DEC3 by Marc FD has extensive deblocking and deringing options (kinda like postprocessing in dshow-filters, only better) among other gadgets. Great solution to gain compressibility. Just great.

Try them and have fun!

@Lothar: Perhaps you care to give us your detailed settings?

[Lothar]

Impos:

Msp: 6 / h263
quantization: 2/4 - 2/6
Two pass: 10/10/20 - 10/15 - bias
No use of altcc.

duration: 22 minutes
size: 200000

[jang0]

Originally posted by Teegedeck
I got no idea what MPEG2dec2 can do.


i think it is by trbarry and it is optimized for SSE2 (pentium 4 processors only)

[Teegedeck]

Originally posted by Lothar
Impos:

Msp: 6 / h263
quantization: 2/4 - 2/6
Two pass: 10/10/20 - 10/15 - bias


Reads a bit like a treasure map. ;) Which value for which setting?
Perhaps you could try high:0 low:0 pb delay:250.
i-frame boost:100 i-frame br-reduction:50 below i-frame distance:10.
no quant limits.

What resolution? How big is the first pass? What filters?

Hopefully someone will have an idea that helps you then.

[digitize]

Personally I think xvid does a good job on anime, some of the unstable builds might jiz things up, but that's why they are unstable builds ;) . First of all, like said before use linear rather than altcc, gives an overall better image. I suggest something like c3d as a denoiser, and warpsharp afterwords. B-frames would give good results (less noise (edit: forgot to add, b-frames also are more efficient than p-frames/compress better. But with the use of b-frames, p-frames are larger)), but recently have been screwing up compressibility (well i think if you just use b-frames on the 2nd pass it's kinda fixed). For quantization anywhere from 2-6, you'll have to play around to get the right number, and quantization type h263. Also i payback proportionally (more of a preference thing). I use tv caps for my source filter to huffyuv and then compress to an mpeg 4 codec, and go for 4 eps a cd (on avg 175mb) with a resolution of 640x480. Well, that's what i use in no specific order, im done rambling now :)