Using Xvid on Anime--> Poor Results

[Dali Lama]

Well, I started using Xvid a while ago and have pumped out 5 good rips now. I like it better than Divx 3 or 4. However, I recently tried to encode an anime and it is not as good as Nandub (Divx 3) by a longshot. I don't know why though. I have tried all of the possible settings.

If anyone has had better encodes with anime or cartoons, inform me of how you did it, thanks.

Dali

[western shinma]

I've used both on anime quite a bit, and the only problem XviD has with animation is more artifacts around sharp edges. However if you look closely this is because it actually makes the edges sharper than in Nandub, due to DivX's prefiltering. So, a smoother filter might help out here.

Nonetheless, there are advantages to using XviD on anime even. Backgrounds aren't as blocky for example, and there aren't any motion trails either. You know, where junk follows something panning across the screen until the next keyframe (not something you can fix with antishit).

Using separate subtitles alleviates the noisier edges problem as well, since that is where it is most noticeable otherwise. Also, make sure you're using H.263 instead of MPEG, this helps a little. Either way, edges aren't not as noisy as with DivX4, which is surprising since that codec is so blurry.

[Dali Lama]

Well Shinma, I have been using the 2d cleaner and warpsharp filters, but maybe a temporal filter will benefit anime as well as movies. Your observations are correct as I have been noticing the same things. However, I do feel that nandub under proper settings will not produce sharp artifacts or that background problem (mostly).

What settings under Xvid do you use (I know there arn't many)?

Thanks,

Dali

[western shinma]

Indeed, the problems are mostly unnoticeable at lower DRFs, but even at 3x they still show up sometimes. It's rather impractical to go around compressing everything at 2x though.

If you're already smoothing, you might try smoothing a bit more, to me XviD looks sharper than Nandub with similar filesizes. It looks about like the difference between neutral bicubic and bilinear on anime.

Still, there don't seem to be any settings to get rid of edge noise completely here either, so maybe someone else will have some ideas.

[Aktan]

I have been testing Xvid on aime as well, but I get motion trials, though I have been using MPEG instead of H.263. The reason why I used MPEG is becuase it was less blocky in gery - darker scenes (Cowboy Bebop Episode 1 Opening) which was a surprised, becuase I expected the same thing H.263 should be smoother. But Divx 4 seems to be even less blocky in the scene and has less motion trials, tho Xvid is sharper.

[western shinma]

What resolution and bitrate were you using in your Cowboy Bebop example?

Can you describe the motion trails in more detail? Is it similar to what happens with DivX 3.x?

[Aktan]

Well i ripped at res of 640x464, aim at the size of 195 MB (233 with audio). The only motion trial I saw was during the title when Spike's plance move toward the right before it shoot all the way to the left (if u know what i mean). I don't know if this happens in 3.11 cause I don't know how to use Nandub really, but 4.12 has little or not noticeable trials and did it pretty well job in the opening scene. Okay looking at it again, the motion trial is very small but still noticable. The reason why divx 4 does better i think is because of the smoothing again.

[manono]

I can only confirm that the plane moving slowly to the right and then shooting off to the left across a medium blue solid background in the Bebop opening song sequence leaves behind lots of motion artifacts in DivX 3.11. I know because I use that scene for testing, and no amount of smoothing using DivX 3.11 can get rid of it. If that's the only motion artifacting you can find using XVid, then I'd say you're a lucky man, and maybe it's time for me to switch.

Aktan-have you tried smoothers (Smart Smoother-2D Cleaner) to clean up that scene? And you might not have the problem at all if you used a lower resolution as you're sort of bit rate constrained at that high resolution.

[Aktan]

The thing is, if i hafta use a filter to fix the blocks, then I rather use divx 4 without a filter (cuase I really don't like picture affecting filters). Divx 4 does not have these block tho on static scenes may contain noticable blocks which can be fixed with a post processing of 1 (0 being the lowest). What I don't get is why MPEG is acually doing better than H. i have just read that external second pass may be better than the internal one. So i'm going to try that since I have basicly been only using the internal one all this time.

[western shinma]

Well, MPEG quantization is supposed to be biased toward low frequency quality, which means it should give less blocky backgrounds but more noise around edges. I'm more sensitive to the latter problem, so I use H.263 instead for animation at least.

[kastro68]

BUT, i think this depends heavily on the bitrate being used. I have found Mpeg only works well for me at high bitrates.

I have recently encoded 25 eps of SMJ and it looks much better with H.263 as opposed to Mpeg quantization.

Good luck with the anime encoding,

and don't forget to share share share

[libredr]

I have problems with MPEG quantizer too, some artifacts appear around edges (black dots) but they are due to a bug in DivX4 post-processing.

[duartix]

Tell me something,
have you tried locking I-frames at quantizer 1?
I'm curious as to the impact this could have on quality of anime sources.
I also suppose that filesize predictability could get a bit loose...

[-h]

Locking the I-frames to 1 seems "unnecessary", but shouldn't affect file size by any large amount (the overflow calc will cater to this).

Try it and see :)

-h

[Aktan]

I have tried it and the file size is WAY off with delta frames looking like total crap. Thought I think I used build 2/17

[kastro68]

@Aktan

I experienced the same problem when i tried to lock the quantizers...atm i am too lazy to muck around with it again and just leave it at the default 1-31 for both I and P frames.


I think before i locked it at 1-6 for I
and 2-6 for P, but i got really bad P frames...can't remember the size though.

[Aktan]

WOW! Build 02/25 fixed almost all of the prblems with the scene I had. And then they just released build 02/26! DAMN! Keep em coming h!

[Dali Lama]

Good News,

I have used Koepi's 26 build with asymetric curve compression and the artifacts in anime are much reduced, good job guys. I agree with Aktan.

Settings:

2-Pass Settings:

Payback Delay: 240
Low Curve Comp: 15%
High Curve Comp: 25%

Search Precision: 6
Quntization Type: H.263

Min Keyframe Interval: 10
Max Keyframe Interval: 300

Min Quant: 2
Max Quant: 8

No Luminance Masking

Debug 2-Pass Options:

I-Frame Min Quant: 2
I-Frame Max Quant: 5
Smooth Quant Checked ON

A serious decrease in haloing or mosquito noise and blocking on static backgrounds.

Oh, btw, I have not experienced any filesize predictabiliy lose like others in the other thread.


Bye,

Dali