XviD sample open for public ridicule

[Hanty]

Been messing about like a madman now doing various 1min clips in XviD. I guess the count has reached around 20 with a lot of tweaking in between. I've had some good looking results but nothing that has made me shift+delete c:\appz\encoding\nandub yet. I will persevere though, I doubt my own abilities rather than the codec at this moment. I present to you this "URL Deleted" for those of you wanting to do some troubleshooting for me.

As you can see the high motion areas look good. No remarks there, but the low-motion, background areas like the ocean and sky suffer from smearing. There's also plenty of macroblocks.

My setup:

2 pass - internal

Desired size : 9216
KF boost : 20%

Curve compression : VBR (ie high/low 0, payback 1)

MSP : 6
Quantitazion type : MPEG
Min/max Keyframe : 10/300

min/max quantitizer : 2/5

I-frame min/max quantitizer : 2/5

You may wonder why I choose VBR, and the answer is that it simply was the best contender along with 15/15 asymmetric. (Tried 25/15, 30/10, 20/10). As you cna see the file became oversized by almost 2 MB which is understandable given my quantitizer settings, but at least I would have wanted a more stable quality then with that oversize. =/

Well, I'm all open for comments on these errors (translation = do my legwork for me). If someone have a tomatoe you are free to throw.

PS. I forgot to mention that I for filters used :
Smart resize (Precise bicubic)
Smart smoother (3/20)
Warp sharp (30)

[-h]

As you can see the high motion areas look good. No remarks there, but the low-motion, background areas like the ocean and sky suffer from smearing. There's also plenty of macroblocks.

Yep, sounds as though you have good high-motion scenes at the expense of low-motion scenes.. just the thing curve compression is around to compensate for ;)

I'd recommend H.263 quantization, no quantizer restrictions, a min keyframe interval of 1 and a payback delay of 100.

Sounds like you're expecting miracles though. If you want the best quality you can get, do a 1-pass DivX5 encode, with quantizers fixed at 3 or 4 and B-frames enabled.

-h

[Hanty]

Hehe, well I dunno what to expect, but it's you people that has been getting my expectations of the codec up up up in here. ;)

[Teegedeck]

Hi there,

a CC of 0 could only be used in Nandub because you could set a pretty high minimum bitrate. And GKnot did that better anyway. :(

So you might want to try GKnot with Xvid if you go for that. ;)

And those strict quantizers really ruin everything that two-pass stands for. :(

Perhaps you should just try not to tweak it and try it with the default values... And, right, use H.263 if you're not aiming for an encode with constand quantizer=2.

[Neo Neko]

I love anime. I have got tons of VHS and DVD. Plus almost 35Gb+ on CD. Using the current XviD decoder the background noise was easy to spot. But it still looked better than many Divx3/Nandub rips I have personally seen. I changed the stream descriptor from xvid to divx and the fourCC from XVID to DIVX. Using the Divx5 render filter with post processing it blows away all Divx3/Nandub encodes I have seen! Especially when it comes to anime. The guys over at super5.com who encoded the DBGT series while they may have done a decent job translating it from cantonese totally screwed the encoding!!!

You have to remember that at this point what you were seeing was Xvid totally unfiltered! Once Nic and the others get postprocessing down in the Xvid decoder say bye bye Divx3/Nandub. If you got any other encodes like that I would definatly be into downloading and possibly replacing some of the less savory Divx3 encodes in my collection.

[morello12]

Hey Neo Neko (or anyone else), I have a test encode that I wouldn't mind hearing your comments on. You can see it at:

here (ftp://anime:ars@137.99.174.201:21000)

I would appreciate comments as well :) I plan on doing the whole series soon, and if anything should be changed I guess now is the time to know. A lot of the pixelation exists on the original Vob files, and IMO, the file linked above surpases them in quality... especially when played through the DivX filter with a post processing of around 4 or so.

But comments are appreciated!

Oh yeah, it was encoded at 640x480 with Xvid using H.263 at around 4500 kb/s I believe, and a few minor filters to clean up the video... but preserve detail. Levels is especially important here. Thanks.

Oh and on topic... ;)

I would like to see a higher resolution personally, unless you're aiming for a small file size.

Also, I would highly recommend this filter chain:

VirtualDub.audio.SetSource(0);
VirtualDub.audio.SetMode(0);
VirtualDub.audio.SetInterleave(1,500,1,0,0);
VirtualDub.audio.SetClipMode(1,1);
VirtualDub.audio.SetConversion(0,0,0,0,0);
VirtualDub.audio.SetVolume();
VirtualDub.audio.SetCompression();
VirtualDub.video.SetDepth(24,24);
VirtualDub.video.SetMode(3);
VirtualDub.video.SetFrameRate(0,1);
VirtualDub.video.SetIVTC(0,0,-1,0);
VirtualDub.video.SetRange(0,5860327);
VirtualDub.video.SetCompression(0x75796668,0,10000,0);
VirtualDub.video.filters.Clear();
VirtualDub.video.filters.Add("deinterlace - area based v1.2");
VirtualDub.video.filters.instance[0].Config(0, 1, 27, 50);
VirtualDub.video.filters.Add("resize");
VirtualDub.video.filters.instance[1].Config(720,1200,134);
VirtualDub.video.filters.Add("Picfix 1.04");
VirtualDub.video.filters.instance[2].Config(6, 4);
VirtualDub.video.filters.Add("resize");
VirtualDub.video.filters.instance[3].Config(720,480,134);
VirtualDub.video.filters.Add("null transform");
VirtualDub.video.filters.instance[4].SetClipping(6,0,6,0);
VirtualDub.video.filters.Add("resize");
VirtualDub.video.filters.instance[5].Config(640,480,134);
VirtualDub.video.filters.Add("filter blender (1.1)");
VirtualDub.video.filters.instance[6].Config (0);
VirtualDub.video.filters.Add("static noise reduction (1.2)");
VirtualDub.video.filters.instance[7].Config (8);
VirtualDub.video.filters.Add("filter blender (1.1)");
VirtualDub.video.filters.instance[8].Config (1);
VirtualDub.video.filters.Add("filter blender (1.1)");
VirtualDub.video.filters.instance[9].Config (2, 50, 50);
VirtualDub.video.filters.Add("levels");
VirtualDub.video.filters.instance[10].Config(0x0000,0xFFFF,0x00FFFE8F,0x1010,0xEBEB, 1);
VirtualDub.subset.Delete();


There seems to be some problems in your encode of background and similiar colored areas remaining in the same area, like the skies when the person is riding the bike. Anyway, pretty good start! Keep comparing the original to the encoded version and adjusting your filters, and it will only get better!

Do not forget to add a levels filter and set the output levels from 16 to 235 or so. This is very important for keeping greys from turning black during playback... and is missing from your chain.

[Neo Neko]

Ouch it is a large one. Damn I gotta wait 8 whole minutes! :D I am getting ready to rip my Knights of Ramune DVD. Still debating wether I should do it as MP4 or OGG.

[Koepi]

Morello encodes the roccosi/movmasty-way?

Somehow this starts to irritate me....

[Neo Neko]

It looks good. But damn it is huge! If the bitrate useage kept constant you would have about 30+ min per 650Mb CD. All I could get out of the AC3 was 2 channel. If it is only 2 channel why use AC3? Maybey I have something misconfigured.

[rui]

Originally posted by Neo Neko
You have to remember that at this point what you were seeing was Xvid totally unfiltered! Once Nic and the others get postprocessing down in the Xvid decoder say bye bye Divx3/Nandub.

What do you mean by that? The xvid ds filter (_Nic's) has post processing for days now.

[Hanty]

If you got any other encodes like that I would definatly be into downloading and possibly replacing some of the less savory Divx3 encodes in my collection.

I'm not sure I'm following you Neo Neko. You should never accept quality of the kind I produced in that XviD clip, it's good for an experimental run with a codec I only have 10 hours or so of experience with, but not to take up your collection with.

If you want to see what I'm shooting for you can check this same clip : "URL deleted" which is a sample from the actual release I did for my group, but in SBC, it has audio subs and all. The full movie was a two CD release, 699Mb et 696Mb.

Also here's a DivX5 sample that I cooked up in three tries with the advise -h gave me, it looks significantly better than the XviD one, but not as tasty as the SBC. However, it's only half the size of the SBC, which is telling something. "URL Deleted"
I'm sure I could get the filesize and quality even better, but this serves good as an example the way it is now. Well, I'm still rooting for XviD however seeing how the DivX5 scene is more concerned about banners than actual encoding. ;)

PS. The server has limited bandwidth, please give some feedback if you download the clips. ;)

PSS. And isn't postprocessing implemented already? (EDIT : nic beat me to it)

PSSS. Superb looking intro morello, I've never seen such a HQ version of Eva. but I've also never seen anyone put 40MB into one minute of playtime. So anything but topquality would be a shame. The sound is alittle funky, some bad spikes, did you use v2a?

[maven]

ever thought about buying the things you want to add to your collection??? doh?!?

[morello12]

Morello encodes the roccosi/movmasty-way?


What exactly does that mean?

It looks good. But damn it is huge! If the bitrate useage kept constant you would have about 30+ min per 650Mb CD. All I could get out of the AC3 was 2 channel. If it is only 2 channel why use AC3? Maybey I have something misconfigured.

Because it's 192 kb/s. If I used any other format, I'd still most likely use 192 kb/s, and I'd rather have the original audio in the clip rather than something resampled.

The sound is alittle funky, some bad spikes, did you use v2a?

That's your decoding, not the ac3. You need to turn off dynamic filters on the ac3 if you're running them. There are no spikes in the ac3.

ever thought about buying the things you want to add to your collection??? doh?!?

Ever thought about posting on topic and not spamming the thread? If you were referring to me, well, you couldn't be more wrong. Either way, I'm surprised to get that crap here. Well, maybe this is why I don't post here regularly.

[Hanty]

That's your decoding, not the ac3. You need to turn off dynamic filters on the ac3 if you're running them. There are no spikes in the ac3.

Oops! My bad, you even mentioned it in the thread.
memo to me : Read entire posts before drawing conclusions.

[The People's Elbow]

@Hanky: A little question...are you comparing the XviD Clip to the SBC Clip you linked? :confused:

[Hanty]

Yes, They are all done by me. What part is it that causes confusion?

[The People's Elbow]

Just because the SBC has an average of about 3600+KBit/s and the Xvid 1860KBit/s...in relation to this circumstance a comparison is sort of useless. Additionally you did the fixed quantisizer thing which isn't that usefull at all ;)

[Hanty]

It's not a comparassion in a way of determining what codec is better or anything like that. I know how to work with SBC, and I don't with XviD. So when you are telling me that fixed quantitizers isn't good then I can only guess what that means.

I would assume that it has to do with my low Quantitizer min/max.
Well I tried with having low min and high max, and it came out looking worse. That's my practical evaluation, care to clarify further?

[The People's Elbow]

care to clarify further?
- NO! ;)
...ok, maybe I'll give it a try :D : As you already recognized the fast motion scenes look well and the low motion scenes show a tendency to get blocky! The problem is that you can't use the Curve Compression with the maximum of q5...well, you can, but it doesn't work, because of the restriction in the Q's! Imagine you have many good looking F-M scenes which already use the Q5 Now you want to apply CC on these setting the % value is set to 20 in order to get better looking L-M scenes and you think the F-M scenes would even look good with Q6/7 or sth. else!
See the point?
The F-M scenes are fixed to the maximum of Q5 and can't give away any bitrate to the L-M scenes...it will probably look worse than without CC as you described!

I hope it's an understandable explanation and that it is correct. (phew, maybe I'm talking crap. ;) )

greetz,
The People's Elbow

[Hanty]

As you already recognized the fast motion scenes look well and the low motion scenes show a tendency to get blocky!

This isn't really what I said, I said the parts in the frame that are in fast motion look good, while in the same frame the parts that are not "involved" with the high motion are looking smeared. These are two different issues.

I mean, the entire clip is just about all high motion. And I experimented by having higher Max Qs without much luck sadly. So, in
your mind, from that setup. What would you change the Qs to?

[The People's Elbow]

Well, I'm no specialist in this case, but as far as I've heard the Default of Min2 MAX31 works nicely, because XviD is very good at deciding which quantisizer at which point gives good picture quality! I guess it doesn't need any quantisizer restrictions!

[Teegedeck]

That's my kinda guy! :)

It's good to read a reasonable statement like this from time to time.

XviD's decisions are indeed very good, and if they were left alone, we perhaps wouldn't need any restrictions. ;) These restrictions kind of _contradict_ the concept of the two-pass and should be used with care. You will need them if you use extreme values on other settings of the codec or mock up the two-pass in any other way. The only instance when I actually _needed_ restricted quantizers was when I did a second pass with a newer version of the codec that had a different inter-/intra-decision (consequently the newer version wanted to set keyframes where the 1st pass' stats only allowed the bitrate for a p-frame and thus tried to compress these keyframes at ridiculous quantizers).

So if you don't want to hurt the curve-compression algo's work, at least set a very high max quantizer that only helps to prevent gross errors; like '12' for a 2-CD and '16' for a 1-CD encoding.

I also agree that a min-quant of 2 makes sense, otherwise the bitreservoir might be emptied on a quant-1 frame.

[Teegedeck]

You know, it can be annoying to tweak for days and then find out that it looks better with standard settings - but it can happen, life isn't always fair...

[The People's Elbow]

Teegedeck wrote:You know, it can be annoying to tweak for days and then find out that it looks better with standard settings - but it can happen, life isn't always fair...
- Oh yeah, that's true...life isn't always fair :D

The People's Elbow wrote:phew, maybe I'm talking crap.
- Examination is the 1st way for improvement! ...why the f**k am I talking about "Quantisizers" all the time...they don't exist!!! That thing is called Quantizer - me sucks ;)

[Teegedeck]

Stop putting yourself down!

I'm sure someone will be happy to do that for you... :)

Seriously, a good post doesn't need excuses for occasional slips in pronunciaiton.

[Hanty]

I volunteer.

Elbow:

You suck! ;)
No seriously, applying those looser settings to the clip now and doing some further experimenting.
Boy am I tired of that sequence.

[The People's Elbow]

Not this time :p

...but check my other posts @ doom9's Forum, I bet you'll find much nonsense and crap inside :D

somehow this is degenerating into a spamming thread :eek: don't blame me...blame canada! ;)

[Teegedeck]

[falling into Monty Python mode:] Aww - you're no fun anymore!

[The People's Elbow]

XviD Experts needed in the XviD Q&A section...stupid Elbow has some questions! :D

[Koepi]

@elbow:

"is ja schon gut, beruhige dich, ist beantwortet." ;P

(translation: "everything's fine, it's already answered, calm down" ;) )

Regards,
Koepi

[Teegedeck]

...now we've all gone so completely off-topic that we'd have to delete a handful of posts (including mine). Let us all try to prevent it from happening again... :)

[The People's Elbow]

@Teegedeck: what happening again? ;)

PS: @Koepi: hmmm, guck nochma' in's Q&A thread :D
(treanslation: "I don't know nothing right now!" ;) )

[Hanty]

Used those "looser" settings now on two clips, and the file came out at 3MB, with a desired file size of 15MB. :eek:

Somewhere there's a spanner in the works, interesting though
for a 3Mb file it looked pretty damn good, I could fit the entire movie on 500MB and still look better than a lot of Flask encodes.

[Teegedeck]

Originally posted by Hanty
Used those "looser" settings now on two clips, and the file came out at 3MB, with a desired file size of 15MB. :eek:


Now, it could be interesting to hear how you calculated that filesize. But seriously, how did you do that, out with it!

[Hanty]

I calculated it like this 15MB*1024 = 15360, if that's what you meant.

CC
payback : 100
High/low : 10/15

Enc tuning
6/MPEG

Q tuning
min/max Qs : 2/12

2-pass
I frame min/max Qs : 2/31

Anyhow, I think I spot where the boo-boo is,
I'll hold my mouth until Iget to hear the word on
the setupfirst though.

[The People's Elbow]

CONFUSION...get's another meaning the longer this thread goes on! ;)
I have no idea what u did wrong :confused:
gimme the boo-boo :D

[saVe]

i had some strange problems too with filesizes.
see http://forum.doom9.org/showthread.php?s=&postid=92225#post92225

i think the codec has some problems with real short clips because i used the settings that caused undersized files on a full movie and everything worked fine.

[Hanty]

Elbow :

High/low : 10/15

That was supposed to be wrong, or so I thought, I found that switching it around to 15/10 didn't change things much.

Alright, now I'm confused, maybe saVe has it right, but that would make experimenting awkward =/

[midiguy]

slightly off topic but seeing as this is about the only place talking about anime encoding in this forum, I figured I would post here.. Is it worth performing IVTC on anime? I have tried both TMPGenc and g knot, and they always seem to make the video jerky at points. Also, they still leave some interlaced frames. What I think is happening is that since anime is not constant 24 fps and can vary, the IVTC is screwing up. it seems like it is removing some "once progressive" frames and leaving some interlaced frames behind. Is there a good way to IVTC anime? or should I just stick the a de-interlacing filter like smart de-interlace? thanks!

[Dali Lama]

I would try out the latest version of Decomb 3.8.

It has a new pattern mode, which should work very well on NTSC, and it has for me.

Just go over to the Avisynth forum and look for the largest thread there...New IVTC or something.

If you like you can use post processing which is a smart deinterlacer if it does miss interlaced frames.

[Hanty]

Don't restrict yourself to one method, use combinations. Personally I've come to the point very I've seen the limitations of just about all ivtc/deinterlace tools. So now I just simply do TMPGEncs auto IVTC and then do a major QC of the result, going back and manually ictv:ing any scenes were the auto has missed. Not something I recommend for the impatient.

[Teegedeck]

Originally posted by Dali Lama

If you like you can use post processing which is a smart deinterlacer if it does miss interlaced frames.

...Just wanted to mention that 'Telecine' in Decomb uses post-processing by default, so you don't need to think about it unless you want to disable it.

[Koepi]

IVTC is related with XviD because...?

SCNR

Koepi