I need some volounteers for a small "blind" test

[Chainmax]

I just took a sample vob from a DVD set I own and encoded it in three different ways. I'd like to know if a few of you guys could download the three encodes (~14.5MBytes each) and tell me which one you think looks better. One thing though: I think I don't have any more bandwith to spare on the server I usually upload my files to, so I'd need some recommendations for a free temp storage server with ~120MBytes capacity.

[trodas]

http://rapidshare.de/

Split per 50MB required, tough... :p

[Chainmax]

Uploading encodes, thanks for the tip :).

[Sirber]

Well, I don't mind hunting squares :D

[Chainmax]

I expect to be able to post all three links in about 1/2 hour. By the way, how can I convert an MP4 file to an MKV one?

[708145]

I expect to be able to post all three links in about 1/2 hour. By the way, how can I convert an MP4 file to an MKV one?

If you upload all 3 files in one archive, we'll be able to download it in less than 3hours. Thanks.

bis besser,
Tobias

[Sirber]

I expect to be able to post all three links in about 1/2 hour. By the way, how can I convert an MP4 file to an MKV one?
Yes. :)

[Chainmax]

708145: I can't upload all three clips to one file because the combined filesize exceeds the 50MB limit.

Sirber: yes what? :confused:

[Sirber]

mkvmerge.exe -o file.mkv file.mp4

[Chainmax]

Merci beaucoup, monsieur :).

Everyone: I'm going to upload one of the clips again, so please be patient.

[Chainmax]

Ok, the files have been uploaded. Here they are:


http://rapidshare.de/files/3858585/1.mkv.html

http://rapidshare.de/files/3859511/2.mkv.html

http://rapidshare.de/files/3862018/3.mkv.html


Remember, no peeking in FFDShow until after deciding which clip (if any) looks better than the others.

[ChronoCross]

Source: http://www.chronocrossdev.com/TestClip.vob
Clip 1: http://www.chronocrossdev.com/1.mkv
Clip 2: http://www.chronocrossdev.com/2.mkv
Clip 3: http://www.chronocrossdev.com/3.mkv

Mirrors for those of you who are impatient.

[stephanV]

I don't have a real preference for any of the videos. Too bad there is no source so I could use VQS ;).

Oh, and for people using mplayer, remux 1.mkv first because display width and heigth are set wrong there. (and mplayer prefers container AR over bitstream AR.)

[Chainmax]

ChronoCross: thanks for the mirroring. I believe you know this clip already, eh ;)?

stephanV: if you want to check out the source sample, you can download it here (http://www.31012.com/~azulftp/TestClip.vob). About 1.mkv, aren't width and height set to 640x480?

Everyone: I really don't know how much bandwith left I have in there so no more than two or three people can download the source.

[Sirber]

Downloadiing at 5KB/s...

[Chainmax]

So much for "rapid"share :(.

[yokem55]

How about setting up a bittorrent of these?

[ChronoCross]

Source Mirrored in above post.

[Chainmax]

Wow, that's going above and beyond the call of duty. My sincerest thanks, ChronoCross :)http://instagiber.net/smiliesdotcom/otn/wink/thumb.gif.


If anyone feels inclined to take a stab at filtering the source, please do so. I've been on and off trying to make a good looking rip of this for many months and never got accaptable results.

[ChronoCross]

Just remember Chainmax. some sources can only be improved so much. If the mastering of the DVD is so horrible that it creates things that cannot be totally filtered then all you can do is make it look it's best and move on.

[stephanV]

stephanV: if you want to check out the source sample, you can download it here (http://www.31012.com/~azulftp/TestClip.vob).
Thx!

About 1.mkv, aren't width and height set to 640x480?

I don't wanna spoil the thing here, but there was a bug in the tool you used causing it to write wrong AR info.

[ChronoCross]

Chainmax: would you like me to re-write the mkv header and post a corrected version?

[Chainmax]

Sure, after you check out all three clips and say which one (if any) looks better.

By the way, I know some things are beyond help, but in this case it's just the blocking I'm going nuts about. I fixed the color bleeding and creeping dots issues and gave up on the sparse abrupt brightness changes (in the closeup of Skinner when he's talking). I just can't overlook the blocking :(.

[ChronoCross]

Sure, after you check out all three clips and say which one (if any) looks better.

By the way, I know some things are beyond help, but in this case it's just the blocking I'm going nuts about. I fixed the color bleeding and creeping dots issues and gave up on the sparse abrupt brightness changes (in the closeup of Skinner when he's talking). I just can't overlook the blocking :(.

you ever used mf's camembert?. it deens the hell out of the source but if used can eliminate alot of source blocking. Also if your using dgindex you can try manao's h.264 deblocking fliter. it works wonders. Check out the readme for instructions.

[Chainmax]

I used Camembert(deblock=true,cleanup=true) some time ago and it made it much worse IIRC. I can't recall if I also had HQDering on the script at the time though.
As for manao's deblocking filter, I had already tried it, but in this particular source FunkyDeblock works better. Some kinds of blocking seems to be better handled by the former and some are better handled by the latter.

[Sirber]

IMO, 1 and 3 are simmilar, while 2 has muskito noise. I did not check, but I suspect #3 to be VP7 while #1 h264.

[IgorC]

Is it blind test? nero h.264, x264 and Xvid build 39 (more probably Xvid 1.1 beta2 or something patched recently).
Source is anime and hasn't details. So all codecs did a good job.

Hm, did you apply same deinterlace for all samples? or you use h.264 + interlace?

[Chainmax]

Yeah, it is a blind test so don't spoil it to everyone else by posting your guesses which are wrong anyway. All three encodes were made from the same avs script. So, you notice no difference at all between the clips?


Sirber: are 1 and 3 too similar to pick a winner? If not, which one do you prefer?


Chronocross, don't forget to express your opinion.


I need a couple more opinions...Sharktooth, Soulhunter, are you guys around?

[Sirber]

Well, I saw some blocks (black) in the first frames in #3, which I don,t remember to have seen in #1. I'd say #1 is winner.

[ChronoCross]

#1 is better. followed by #3 then #2

[Sirber]

1 > 3 > 2

I agree :)

[Chainmax]

Interesting, I hope some more votes come up. Thanks for the collaboration guys :).

By the way, a new filterchain is going to be tried shortly, wish me luck :).

[Chainmax]

OMFG, I think I found a way to remove pretty much all the blocking while staying at 640x480 and without leaving a nasty blur of a picture. It was in front of me all along :). As soon as I get a couple of extra votes, I'll post a version for you guys to check out.

[DeathTheSheep]

Oooh, nice. I, for one, dispise blocking. On this particular source, I'd say 3 and 1 tie in nicely, leaving 2 behind. Those blasted ASP codecs of yesterday.... :p

And about rapidshare... Don't use it. Use something else like www.megaupload.com (250MB limit) or www.nerdswap.com (2GB file limit, limited time) or even www.yousendit.com (1GB, limited downloads+time).

Rapidshare seems to be rather limited compared to others (and overworked, lol).

[Chainmax]

Three opinions that say the same thing are good enough. Here's the skinny on each file:


1.mkv: x264 Rev 281A --> 3 reference frames; 2 bframes (pyramid, adaptive, weighted); deblocking=2,2; CABAC; RDO; Chroma ME; All macroblock options; Uneven Multi Hexagon search.

2.mkv: Koepi's Xvid v1.1b2 --> H.263; Cartoon mode; Chroma optimizer; VHQ4, VHQ for bframes; Trellis.

3.mkv: Nero Recode (NVE-3.1.0.14a) --> Decision Quality High; Maximum vector range = -512 to 511.75; 3 reference frames; 2 bframes; All encoding tools with non-adaptive deblocking set @ 2; Chroma optimization; high psycho visual enhancements.


In each case, non specified parameters were left at default settings.

[Chainmax]

I couldn't manage to complete an upload to NerdSwap, and both yousendit and megaupload only allow uploading to an e-mail adress. So, through RapidShare again, here is the latest version of the clip:

http://rapidshare.de/files/4016334/NewTry.mp4.html

[Inventive Software]

At a guess, Nero Recode is MPEG-4 ASP, XviD we know is ASP, and x264 we know is AVC.

With the settings you have with x264 it's no surprise that it's better than XviD, though quite surprising that it's similar to Nero Recode.

Do you have encode times for each? Or is that not the point? :D

[yokem55]

Well, as the nero recode config mentions reference frames, I would presume it to be the Nero AVC codec, not the ASP one.

[[)370|\|470!2]

3.mkv=AVC, but correct me if i'm wrong, ND wouldn't allow cfr, only vfr afaik.
So, Chainmax, how did you managed to create cfr video in nero? o_0

[Sirber]

OMFG, I think I found a way to remove pretty much all the blocking while staying at 640x480 and without leaving a nasty blur of a picture. It was in front of me all along :). As soon as I get a couple of extra votes, I'll post a version for you guys to check out.
Where? I'm interrested! :D

[Chainmax]

Sorry about the lack of answers, I forgot about this thread :o.

Inventive Software: like yokem55 said, I used Recode's AVC encoding.

[)370|\|470!2: I don't know what you're talking about, I just loaded the avisynth script in Recode and encoded.

Sirber: it's nothing revolutionary, I just removed LimitedSharpen from the script. Apparently, if the source is has bad enough blocking then standard sharpening will result in visible blocks no matter how much deblocking and smoothing you do (for this clip, at least).

[Sirber]

Well, your miracle way was to filter less? LOL

[Chainmax]

Actually, just removing sharpening but yeah, you're right :o :rolleyes: :p.

[DeathTheSheep]

Wow. You gotta love this guy :d

Yeah, the same is true of resizing stuff for ASP codecs-- don't use a super-sharp resize if you want good compressibility (and therefore less blocks). No, instead use the simple old precise bilinear and encode with no psychovisual enhancements. Good stuff if you don't mind the lack of extra crispness. Cheers

[ggab]

@Chainmax, could u post both TestClip.vob's Avisynth script and x264 1.mkv's command line?

thanks, GABriel from Argentina (latest match: ARG 0 - URU 1 ;) )


maybe something like this, no? (from "x264: extreme quality on anime!"'s thread)
MPEG2Source("X:\wherever\SimpTest.d2v",cpu2="ooooxx")
DeDot()
TFM(d2v="X:\wherever\SimpTest.d2v",mode=3,PP=7,mChroma=true,chroma=true,mi=50)
TDecimate(mode=1)
FixChromaBleeding()
LumaYV12(lumoff=-2,lumgain=1.0)
ColorYUV(levels="pc->tv")
Deblock()
Deen()
HQDering()
Crop(12,0,700,476,align=true)
aWarpSharp(depth=16,cm=1)
BicubicResize(640,480,0,0.5)
BlindPP(quant=0,cpu=0,cpu2="ooooxx")
LimitedSharpen()
and:
x264 --pass 1 --bitrate 1620 --keyint 240 --min-keyint 24 --bframes 3 --b-pyramid --ref 8 --filter -6:-6 --analyse all --weightb --me umh --merange 16 --subme 5 --8x8dct -o NUL x264Try.avs

x264 --pass 2 --bitrate 1620 --keyint 240 --min-keyint 24 --bframes 3 --b-pyramid --ref 8 --filter -6:-6 --analyse all --weightb --me umh --merange 16 --subme 5 --8x8dct -o Pruebax264.mp4 x264Try.avs

Saludos!

[Chainmax]

Sorry for taking so long to answer, ggab. I must have encoded that clip over a hundred times (counting different codecs) and I never was able to eliminate the blocking in the green background from the scene where Homer and Marge try to find their seat. The source itself must have the blocks handdrawn on it :scared:. Anyway, the latest filterchain I tried was the following one:

DeDot()
TFM(d2v="X:\wherever\SimpTest.d2v",mode=6,PP=7,slow=2,mChroma=true,chroma=true)
TDecimate(mode=1)
FixChromaBleeding()
org=last
LumaYV12(lumoff=-2,lumgain=1.0)
ColorYUV(levels="pc->tv")
LRemoveDust_YV12(17,2)
FunkyDeblock() #or Deblock(quant=30), whichever you prefer
aWarpSharp(depth=16,cm=1)
MergeChroma(org)
Crop(12,0,700,476,align=true)
#Lanczos4Resize(640,480)
BicubicResize(640,480,0,0.5)
HQDering()
#LimitedSharpen()

The Lanczos4Resize and LimitedSharpen line are commented out because by sharpening the picture they worsen the blocking. If I ever find a way to eliminate said blocking then I'll use them. I removed the BlindPP line and the cpu2 parameter in the MPEG2Source line because HQDering and aWarpSharp seem to remove all the haloing by themselves (yup, aWarpSharp can also slightly weaken halos).

As for x264 options, I don't use straight CLI commands since MeGUI is so much easier. Here are the settings I recommend:

Keyframe Interval = 10xFramerate
Reference frames = 5 (or 16 + Mixed)
BFrames = 1 (or 3 + Adaptive + Pyramid + Weighted Prediction)
Deblocking = 0,0 (up to 2,2 if needed, though I'd prefer using denoising instead)
CABAC
Subpixel Refinement = RDO (or RDO level 2)
High Profile with all Macroblock options enabled
ME Algo = uneven multi hexagon
Min GOP size = framerate
Custom matrices: not really needed, unless you go for ultra low or ultra high bits/pixel*frame scenarios.

I can provide links to the needed items if you can't find them.

[QuadraQ]

Without looking at the other messages I would say that I thought the quality from best to worst was like this:

2
3
1

(I'm doing this blind, I never looked at what they are at the time of the post.)