XviD in HDTV at very high bitrates: bug?

[Hironimo]

Hi there.

I've been trying around getting a HDTV source (1920x1088) as good as possible on a DVD in XviD. The target bitrate is about 6500kbit/s to fit it plus 448kbit/s AC3 sound. To reduce blocking, CPU usage and because most displayes cant handle such high resolutions I resized it to 1280x720.

However, I noticed the bitrate never reched 6500, it always remained at about 6000, and even changing the target bitrate to 10000 would not change it in the least. And this is not due to the video already having been encoded optimally - there is still some blocking to be found - the Compare function of avisynth rendered me an overall PSNR of 44.7529 (already having tried around with some optimisations), which is far from the best.

So I now wonder - does XviD have a bug preventing higher bitrates then about 6000kbit/s?

Here is my data:
- using XviD 1.0.2 (koepi's latest stable build)
- I think I'm using XviDs own decoder (no other installed as those somtimes tend to play the video choppy)
- I still have to try out if it happens with other sources
- My System: P4 2.5GHz, 1GB Ram, Radeon 9600 Card, Audigy 2 Platinum, Windows XP, usually using MediaPlayer 6.4
- XviD options: Unrestricted Profile, tried various quantization types, deactivated: Packed bitstream, Adapative Quantization, Interlaced Encoding, Quarter Pixel, GMC, Reduced Resolution; tried VHQ mode both 1 and 4, Quantization: Min quantizers at 2, Trellis activated. (Note that I have deactivated all the Optimizations like QPEL and GMC because it takes more time to compress and decompress and sometimes tend to have problems in playback, specifically on stand-alones. I tried both single and 2-pass, where the latter resulted in a bit smaller files and slightly better PSNR.
- I'm using Avisynth to prepare the HDTV data, but as the script is a bit long I will only post upon request. It seemes fine with VirtualDub.
- I'm encoding to AVI, usually using one of the latest VirtualDubMod compiles, AC3 is added in another step after the video encode finished.


I would be glad if you could help me out!
Hironimo

[Versa]

Beats me, I have never bothered to get up that high in bitrate. I always specify target size, not bitrate, but my avg 1280x720p hdtv encode bitrate is around 3000, which looks great to me and a handful of other people that I give it to.

You ARE doing two pass encoding right?

[Hironimo]

I am using such high bitrates because I am not satisfied with bitrates as 3000kbit/s - very noticeable blocking, and as I want to save it to DVD anyway, why not use the full capacity of it?

And, as mentioned now after adding it to the first post, I used 2-pass, but also tried single-pass. It really does not matter which I use concerning the problem - 2-pass gets slightly better quality and leads to slighly smaller files - in a one-minute encode 45MB instead of 48MB.

I'll try 3200kbit/s encode now to see how the quality is, again to have something to compare.

Hironimo

[Manao]

Quantization: Min quantizers at 2I bet all your frames are encoded at quantizer 2 : you're saturating your bitrate, and you can't go higher than 6500 on that particular video. Try a more difficult video, you'll see bitrate will raise.

[Hironimo]

Last time I checked the quantizisers were at 4-6 average when encoding. And if I really have saturated the bitrate/gave him something too easy to encode, why would I then still see some blocking?

But I will select a more difficult part of the video to encode (encoding all of it would take too much time, even a minute of video takes me 10-20 minutes) and get back with results.

Hironimo

[Teegedeck]

I have encoded 720p HDTV with XviD 1.1, using the SixOfNine matrix at quants 3 or 4 (before there was VBV). At quant 4 I reached an average bitrate of about 6 Mbps and peaks of about 24 Mbps, at quant 3 it averaged around 9 Mbps and peaked at about 29 Mbps. For exact values I'd have to check at home.

I used Matroska-Property's bitrate function to get those values.

So at leat in 1.1 there does not seem to be a bug of any kind that would limit the bitrate.

[Hironimo]

With qant 4 and 3 do you mean you encode quality-based, not bitrate-based? If this is the case, could you try bitrate-based if you can reach such high bitrates (at least clearly exceeding 6000kbit)?

You mentioned Matroska-Property's bitrate function - can you explain how to get it? Is it part of the matroska pack? If so, where to find it after installing it? (Sorry, I have no experience with Matroska yet)

Hironimo

[Teegedeck]

Hmm, I don't know when I'll have time to do it, but I'm gonna check. I'm just experimenting with HDTV, not seriously backing it up.

MatroskaProp should be in the Matroska Pack Full (as an option during the installation). Right-click on the mkv, select the file properties, in the properties go to the tab Matroska Info (or what it was called) and click on the 'bitrate' button. It'll take a while...

[Hironimo]

Thanx.
I'll try. Does it only work for .mkv files or also for .avi and maybe .ogm?

[Teegedeck]

Only mkv. It is called MatroskaProp accordingly.

[Hironimo]

I've installed matroska (full pack version 1.0.3), used direct stream copy to convert from avi to mkv and tried to look at the bitrate, but that causes an unhandled exception and it seemes to fail to generate debug information (though when a window comes allowing me to debug I don't do that as I assume I would debug assembler code, and and that only makes sense for a developer of matroska).

I just did more encodes with a section that seemes to have more motion and did encodes at 6500kbit and 10000kbit, resulting in an average of ~5693kbit and ~8224kbit.
Notice though that these values come from single-pass encoding.
I did a two-pass encoding at 6500kbit/s and it leaded to 6258kbit/s average. One thing I did not yet notice that kbit in divx is 1000 bit, not 1024 bit as I assumed and calculated. An oversight of mine,
I'll have to ajust my calculations when I specify values for target bitrates (which I tend to measure in kbyte/s where 1kbyte is 1024 byte).

So you are half right.
It is possible to encode at higher bitrates then I originally feared, but it encodes at lower bitrates then specified, which would be ok if the quality is near-perfect, but it is not - there are still blocks when looking close. Would'nt it be te aim to encode as good as possible while staying at the desired bitrate and only do a lower bitrate if there is no gain in quality?

Hironimo

[Mug Funky]

are you using the MPEG matrix or H-263? the latter is basically a flat matrix, so it'll block as much as it rings (which is not much at all). switch to MPEG, or a custom one and bitrate will go up (and blocks might decrease for any given quantizer, but i'm not sure).

i haven't really played with HD stuff, but i'm pretty capable of getting high bitrates on regular stuff :)

if it still ends up too small, try putting "limitedsharpen" in your script :) that'll inflate the bitrate some (but will look pretty nice).

what sort of material are you encoding (aside from HD). natural images i suppose.

[Hironimo]

I tried h.263, mpeg and custom matrix, in all it is the case that bitrate is lower then specified. I encode wildlife-video - there is a bbc production that recently on Pro7 - a german private channel, and they premiered with hdtv in this video called "Pride".
I think mpeg gives a bit better results, but custom might be a bit too much. I will give sharpen a try, but the image is pretty sharp as it is.

[Manao]

Hironimo : what's the length of your clip ?

One pass is not meant for bitrate accuracy, but the only cases - which I know - in which two-pass doesn't reach the target bitrate are :

* bitrate saturation, because of saturation at quant 2 -> undersizing
* use of quant 1, which often misleads the RC algorithm if its overflow settings are too loose by default for quant 1 -> oversizing or undersizing.

You're not in the second case, so it would seem that the first is happening. I checked with a sequence I upsized and sharpened to 1280x720, with low & fast sequences, and XviD 1.0.2 and XviD 1.1 did reach the target bitrate. The max average bitrate reachable on that sequence was 8250 kbit / sec.

[Blue_MiSfit]

I suggest using a higher quality matrix - such as Sharktooth's EQMv3 series...

Perhaps his "general purpose" matrix, or even the higher bitrate one (don't recall the name off the top of my head) would be best suited to your case.

Also, I would reccommend perhaps enabling quarterpixel, as this seems to "sharpen" the image a bit to me, and can also increase bitrate.

If you want to see the very best the codec can do with your settings and input file - then do a single pass constant quality encode at quantizer 2. This will put your I and P frames at Q2, and your B frames at Q4 (If you have bframes at defaults anyway).

I think what the problem here is that you're a little too attached to this idea of bitrate dictating the quality of the movie. Let go of worrying about bitrate, and worry more about hitting a target filesize, and the quantizers therein involved :)

Just a humble suggestion
~misfit

[Koepi]

1pass cbr isn't the most accurate solution in xvid. Try proper 2pass (best) or 1pass at fixed quantizer. That'll lead to a much better quality.

You observed that yourself already.

I think either your source or(and) your avisynth script are "guilty".

MPEG2-streams that are broadcasted with DVB-* (worst case: DVB-T) are blocky already! If you take the time to watch the source without post processing you'll most likely find blocks at the same spaces where they are in the final xvid.

With some bad DVDs this effect has happened to me, and I thought I messed up something. But it was the source. Try using mpeg2source with postprocessing ( like mpeg2source("mymovie.d2v", cpu=xxxxxx) ).

I hope this helps

Koepi

[Didée]

Lately I worked on the very same HDTV stream (Pride). It's only that I did no full encoding, since my capp'ed stream was b0rked hoplessly in too many places.

For producing some snips, I used the following script:
mpeg2source("Pride.d2v", cpu=2, moderate_h=40, moderate_v=50)
crop(0,0,-0,-8)
undot().blur(.3) .temporalsoften(2,4,6,23,2)
LimitedSharpen(ss_x=1.0,ss_y=1.0,dest_x=960,dest_y=720,Smode=2,strength=100)

Note that this is "standard-doesn't-exist" anamorphic 720p. But before bashing me, you first should try it out yourself ;)

For encoding, I used a modification (cheaper Intra coding :p ) of the EQM-v3HR matrix @ quant 4. ...Or was it q5? I'm not completely sure right now :o

XviD settings were standard, plus
- custom quant (obviously)
- qpel
- Bframes 2/1.25/1.5 / sensitivity 4
- vhq-4 plus B-vhq
- trellis

This gave me a stream of ~5500 kbps, which looked pretty nice as long as the input stream wasn't already destroyed. To quickly get 6000 kbps, use CBR=6000~6200, and cap quants to [3,4]. 2-pass would be better of course, but that takes so damn long with HDTV input :(

BTW, while the transmission had good overall quality, to my eyes it didn't offer that kind of definete sharpness that HDTV @ 1920 could offer. That's why I pushed it a little with LS(), and the result seemed fine to me :)


edit: corrected parameters to the real ones.

[guada]

Hironimo, can you tell me the approximate time of your encoding of 1920*1088 with a bitrate of 6500 on 1H30 movie?

[Sharktooth]

Originally posted by Didée
For encoding, I used a modification (cheaper Intra coding :p ) of the EQM-v3HR matrix @ quant 4. ...Or was it q5?
Can you PM me your modifications?

[Hironimo]

@Manao
The overall length is about 88 minutes, but I am currently just testing with encodings, which are once minute in length.

First off, if you are speaking of quants in terms of quality encoding,
I never ever did encoding via quality setting because of the unpredictable target size.
As to the minimum quantizier settings: I both tried standard and minimum of 2. First gave me a bit better results in quality I think
but the average bitrate did not vary much.
The first point could indeed be the case, as quant 2 is always very high in the encoding status window.

What exactly do you understand as "bitrate saturation"?


@Blue_MiSfit
I tested with Didees SixOfNine Max=24 Matrix. I cannot say for sure but it resulted in a decrease of bitrate and also quality - though barely visible - i measured it with avisynths Compare.

I tried quarterpixel once and it requires much more processing power, both in encoding and decoding, which is already very high with hdtv video. It seemed to get worse results then without also, quality wise.

Sour suggestion of using quality encode is worth a try but I hate that I cannot predict filesize then.

As to me being too attached to the idea of bitrate dictating quality I can understand but I in fact look on resulting quality more then in resulting filesize - in the recent tests I did comparisons to the original source always do determine the quality.

You say I should worry more about hitting target filesize. But is'nt target filesize calculated by average bitrate multipiled by time in seconds?
This is why I doo look on the bitrate. I calculate what maximum bitrate fits into the target filesize and then try to get as close to that bitrate as possible.
Is'nt it true that a higher bitrate yields better quality in most cases (always assuming that encoding has been done with settings approaching optimum)?


@Koepi
I agree that 2-pass is better then 1-pass cbr (well, there are always exceptions, i've encoundered one).

I have checked the source: seen where there were blocks in the resulting avi and checked if they existed in the original. They did not.
The source is a good 19mbit mpeg2 stream. I did observe minor blocking in the source at least at one point, but I doubt it has much effect.
I would have given a link to a part of the stream, but did not yet manage to extract it properly, hdtv2mpeg2 (1.10.6) seemes to have bugs (or I do not know how to use it ^-^). The whole stream is however available for download in eMule, search for "ProSieben HD Pride" - it will take an awfully long time to download though. I'm not finished myself fully. (I would have captured myself, but unfortunately its a heck of an effort to get a sattelite-dish allowed where I live :()
I though of adding postprocessing but I think it also reduces sharpness in addition to "rounding" blocks (and it's slower).
But just to compare i'll do a test encode with postprocessing and see how it affects bitrate.


@Didée
Nice to know someone else here knows this video.

Why do you do undot, blur and temporalsoften?
Filters like this would be ok for analog reception or noisy dvds,
but at least in case of Pride I could not detect any noise (well, some scenes have some but this came straight from the camera recording it I guess).

>Note that this is "standard-doesn't-exist" anamorphic 720p. But >before bashing me, you first should try it out yourself
Well, 960x720 is a bit strange but if it fits you :)) Do you correct the aspect ratio while playing, then?

How does "Bframes 2/1.25/1.5 / sensitivity 4" affect encoding?

With "cap quants to [3,4]", what do you mean?
Which settings did you change to what?

I did not do with sharpen. Thats my script:

loadplugin("E:\Programme\AviSynth 2.5\plugins\dgdecode.dll")
LoadVirtualdubPlugin("K:\pride_hdtv\delogo\delogo.vdf","DeLogo")

source=mpeg2source("K:\pride_hdtv\ts\teil1.d2v",idct=5).crop(0,0,0,-8)

firstpart=source.Trim(1178,1448).Lanczos4Resize(1280,720,240,133,1440,815)

secondpartorig=source.Trim(1659,0)
logo=secondpartorig.crop(1680,0,0,240)
logo=logo.blindpp(cpu=4)
logo=logo.converttorgb32()
logo=logo.delogo(1, "", "K:\pride_hdtv\delogo\delogo-deblend3.bmp", "K:\pride_hdtv\delogo\delogo-alpha.bmp", \
"K:\pride_hdtv\delogo\delogo-color.bmp", "K:\pride_hdtv\delogo\delogo-repair.bmp", 80, 31, 0, 11)
logo=logo.converttoyv12()
left=secondpartorig.crop(0,0,1680,240)
bottom=secondpartorig.crop(0,240,0,0)

secondpart=stackvertical(stackhorizontal(left,logo),bottom)

firstpart+Lanczos4Resize(secondpart,1280,720)


@guada
Its really hard to estimate because there are so many variables:
- which avisynth filters you use
- which XvidD options you use
- at which resolution you encode
- how fast your system is.

But to give you an idea:
On my system, with bare minimum filters at original resolution (no rezize), just the bottom 8 pixel crop, I estimate 32 hours.
With all the filters I use (see above) I estimate 17,6 hours.
This is single pass with qpel and others off but wide search.
But as I could now only test via remote desktop I do not know how valid those numbers are, I would have to try again after a reboot directly at the pc (as mentioned P4 2.4GHz).
This shows that the filters do not take as much as encoding at full size.

Cheers, Hironimo

[Manao]

By bitrate saturation, I mean exactly what you're observing : almost all the quantizers are '2'. When that happens, it means that ( providing you don't allows the use of quant 1 ) you can't reach a higher bitrate than the one you're getting.

[Sharktooth]

Uhm... try with EQM V3EHR... it's an insanely high bitrate matrix.
I doubt you can saturate it...

[Hironimo]

@Didée
>to my eyes it didn't offer that kind of definete sharpness that HDTV @ 1920 could offer

I have seen about half a dozen hdtv sources by now (ok, not much), and I thought the sharpness to be quite ok. Sure technically more is possible (like the sharpness you get when you have a hdtv source and downsize it to DVD resolution with Lanczos(4)), but still I am satisfied. I think when its unsharp its mostly the lens, not the encoding.

http://81.223.129.74/pictures/prideshot1.jpg
http://81.223.129.74/pictures/prideshot2.jpg

Hironimo

[Hironimo]

@Sharktooth

I'll try your matrix.

@Didée

I tried to use your LimitedSharpen (created a file LimitedSharpen.avsi in the avisynth plugin directory and put in the script posted, but DEdgeMask is not available and the MaskTools are not available for download because the site is offline.

Hironimo

[Manao]

the MaskTools are not available for download because the site is offline.Yeah, sorry for that, but my ISP seems to have a lot of problems. If meanwhile someone could put a recent ( > 1.5.1 ) version online, I woudl be grateful.

[ARDA]

here you have

http://www.iespana.es/Ardaversions/MaskTools-v1.5.4.zip

[Manao]

Thank you ARDA.

Now that I'm back home, I can check and restore my website more easily. But I really need to find something more reliable :(

[Andrey]

>>MPEG2-streams that are broadcasted with DVB-* (worst case: DVB-T) are blocky already!... With some bad DVDs this effect has happened to me...
Yep. Sometimes it is.
I get StarWars II copy (PAL) with extreme blocks. It is completely not compressable by XVID :(
BTW, about pictures posted - first one is extremely blocked too.
It can possibly give unpredicable results after reencoding, IMHO...

[Soulhunter]

Filtered and downsized it looks acceptable...

http://img101.exs.cx/img101/8869/1024.th.png (http://img101.exs.cx/my.php?loc=img101&image=1024.png) http://img101.exs.cx/img101/6750/1155.th.png (http://img101.exs.cx/my.php?loc=img101&image=1155.png)


Bye

[guada]

Hello Soul,

I appreciate the definite pictorial, hot and better quality.
I think that you can make again better, no.:p

Bye ;)

[Hironimo]

@Andrey
>BTW, about pictures posted - first one is extremely blocked too.

I do not concider that blocky.
I concider THIS blocky:
http://81.223.129.74/pictures/prideshot3.jpg

@Didée

I got to try your LimitedSharpen filter successfully now,
and when looking at the things that should be epecially sharpened, like edges or features, its great, but it does not look so well on more smeary parts as happens with captured motion.
I'll stay without it for now.

@Soulhunter
For some strange reason your matrix produced really weired and blocky results.
These only show up when I play it with a player (windows media player 6.4) - it looked ok in VirtualDub.
http://81.223.129.74/pictures/prideblocking.jpg
The quants did look unusual too:
http://81.223.129.74/pictures/EQM_V3EHR_XvidStatus.PNG

Also, you told that it is unlikely to saturate the bitrate with it,
but the resulting filesize is the smallest of any test encodes yet
(41Mb and not the usual 45-49mb).

Can you tell me what might be causing this?

Hironimo

[Manao]

Bad looking in Media Player Classic is due to the use of a decoder which can't handle the matrix you used ( matrices with up left coefficient which is 8 ).

Now, for the quant distribution, it's a normal quant distribution, nothing wrong with it.

For the size : how did you test it ? One pass average bitrate ? Then it's normal. Such matrix can highly disorganize the RC, i think, because frame size at low quant are very high.

[Soulhunter]

Originally posted by Hironimo

For some strange reason your matrix produced really weired and blocky results...
Read Manaos answer...

Try to use ffdshow for decoding !!!

Btw, which of my matrices have you used ???

Originally posted by Hironimo

http://81.223.129.74/pictures/EQM_V3EHR_XvidStatus.PNG

Also, you told that it is unlikely to saturate...
Are you still talking about my matrix ???

Coz, the picture name says its EQM V3EHR...

Anyway, it cant be saturated as there are no 1/2/3 quants !!!

If you would send me the source (or a part of it) I would like to do some tests... :)

Originally posted by guada

Hello Soul,

I appreciate the definite pictorial, hot and better quality.
I think that you can make again better, no.:p

Uhm, what do you mean with this ???

Your post mystifys me... :D


Bye

[Hironimo]

@Soulhunter
Yes, I did use your matrix.
What I find strange about the screenshot is that when encoding with default matrices it will use 1,2 and 3 quants, but it will not with your matrix.

I had some trouble extacting a portion of a .ts file.
Firstly it remembered include/exclude selection even though I deleted all .tri files, secondly I could not save it to .mpg - the result would be a file only a few kb in size.
It worked when exporting to .ts files though. Do you have any idea why this might be? I used HDTV2MPEG2 1.10.6.

I have done two selections.
The first is from the beginning of the capture (you can use the script I posted before directly on it), the second removed some of it focusing on a more interesting part.
I'm still working on adapting the scripts for you now though.
You can get them at ftp://hdtvtest:pride@81.223.129.74/

Hironimo

[Soulhunter]

Originally posted by Hironimo

What I find strange about the screenshot is that when encoding with default matrices it will use 1,2 and 3 quants, but it will not with your matrix.

Odd, there must be something borked @ your side !!!

Im able to produce much higher bitrates, look...

http://img66.exs.cx/img66/2820/HDSHv3.png


EDIT: I cant reach the "ts" folder (no permission) !!!


Bye

[Hironimo]

I'll try around again.
But I don't think i'll use a customized matrix in the end cause I want the video to be playable without hassles (somone not so adept in encoding might have a hard time finding out why its not playing correctly).

I made an additional section 1b where I removed an ad before the video and uploaded addapted scripts. You have to ofcourse ajust the paths in the .avs and .d2v file.

Permission should be ok now.

Hironimo

[Sharktooth]

Hironimo, maybe you used MY matrix... EQM V3EHR.
BTW, im downloading your sources to see what i can do but i'm in a PAL country and never had any experience with telecined materials...

[Hironimo]

I'm from Austria, and we have PAL too (and I like it!)
So no fear about interlacing or teceline - I dont have real experience there either :)

Ofcourse I used your matrix. I downloaded the last link of customized matrixes that show up with every of your posts.

Should I have used another one?

[Sharktooth]

...you can try also EQM V3UHR. It's still a ultra-high bitrate matrix :)

[Soulhunter]

Uhm, I load the script...

http://s04.imagehost.org/0220/01_JPG.t.jpg (http://s04.imagehost.org/view.php?image=/0220/01.JPG)

...and get this mess !!!


Bye

[Hironimo]

That is strange.
To be sure it is not due to incorrectly transmitted files i've uploaded md5sum.exe and md5sums.txt. Call md5sum --check md5sums.txt to be sure all is ok.

If it is then maybe you should run DGIndex.exe (in dgmpgdec1012a.zip if you need it) again to generate the -.d2v file(s) again to be sure this is ok for your configuration.

If it does not display correctly already in DGIndex.exe I am out of suggestions for now. If it is ok in DGIndex but not in the avs script then maybe simplyfy the script to the bare minimum and see if that corrects it.

It would be good to narrow down the problem-causer.

Hironimo

[Soulhunter]

EDIT: Update...

Im still not able to decode the stream...

http://img66.exs.cx/img66/3830/DGindex.png

Dunno why, but I get only a black picture !!!

As info for sample1b I get 2223 frames, but only 0.770fps...

Havent worked with ts files before, maybe someone else has a clue ???


Bye

[Hironimo]

Ah. I might know why that is.
You have to tell DGIndex the PID's of the video and audio stream.
Go to Stream->Set PIDs and enter 134 for Video PID and 100 for Audio PID.

To detect the PIDs of a .ts you can use hdtv2mpeg2 (version 1.10.6 or later) or better Project X (hdtv2mpeg2 doesnt detect the audio pid correctly for this .ts).

Tell me if that solved your problem.
Hironimo

[Soulhunter]

Yeah, now it works... :)


Thanks n' Bye

[Soulhunter]

Ok, it seems there is really something wrong at your side !!!

Even with denoising I got 9761kbps @ the first pass...

http://img66.exs.cx/img66/6081/SHv3_1stP.png

So, I did a 2pass encode @ 5500kbps with Soulhunters v3 !!!

And considering the relative bad source, the result is imo nice...

http://img66.exs.cx/img66/4893/0549.th.jpg (http://img66.exs.cx/my.php?loc=img66&image=0549.jpg) http://img66.exs.cx/img66/4364/11213.th.jpg (http://img66.exs.cx/my.php?loc=img66&image=11213.jpg) http://img66.exs.cx/img66/9509/11217.th.jpg (http://img66.exs.cx/my.php?loc=img66&image=11217.jpg)

Next time I should throw some pp @ the source !!!

Btw, I have forgotten to use DeLogo as well... :D


Bye

[Didée]

Oh boys. Me feels like talking to the wind. Still no idea about the reasons for my "strange" handling of the source, no? (And that was only a single shot from the hip.)

BTW, I'd be a little curious about what bitrate eqm-v3uhr would return ... /*tries to make an earnest face*/

[Aktan]

Clip Prosieben HD Pride Sample 2.ts

Avisynth Script:

#Renamed from DGDecode.dll
LoadPlugin("D:\PROGRA~1\GORDIA~1\mpeg2dec2.dll")
LoadPlugin("D:\PROGRA~1\GORDIA~1\ColorMatrix.dll")
mpeg2source("Test.d2v", IDCT=7,CPU=4)
Lanczos4Resize(1280,720)
ColorMatrix()

Koepi's 1.0.2 binary

Profile: Unrestricted
MPEG Matrix
Full quality first pass checked
VHQ-4
Max I-frame interval: 250
Trellis quant
Rest is default

1-pass:

http://img84.exs.cx/img84/5682/1-pass.png

Overflow control strength: 0

2-pass:

http://img84.exs.cx/img84/1726/2-pass.png

If you want the 2-pass clip, PM me Hironimo. I don't got a lot of bandwidth to share with everyone, sorry.

[Soulhunter]

Originally posted by Didée

Oh boys. Me feels like talking to the wind. Still no idea about the reasons for my "strange" handling of the source, no?

Uhm, Im not really sure what you mean ???

And Im not even sure if you mean me... :D

Btw, have you solved the last eye-riddle ???


Bye

[Hironimo]

Hi there.

I don't know why the bitrates were less before, but now,
after having finished downloading all .ts parts and going
over to the final encode the bitrates appear much better:
http://81.223.129.74/pictures/finalencode.PNG
This is single-pass at 6600kbit/s btw, two-pass is just
too long to wait for with the rather subtle quality increase.

Well, sometimes video encoding is a mystery...
Was it just becaue one minute of test encoding was
not enough of because the selection was not well
chosen, or something else?

I'm sorry if i've caused confusion.
I'll maybe report with encoding results.

Hironimo

[Aktan]

Is it me, or is the quant range huge? If you don't have the time to do 2-pass, at least limit your I frames to 1-5, P frames to 2-6, and B-frames to 3-7. With a quant range of 1-14, I bet there are definatly scenes where the quality is great and scenes where the quality is real bad.