QuEnc requests/ideas [Archive]

Paced

7th November 2004, 23:23

Originally posted by burnman99
The ability to enocde audio & video separately & save as .mp2 & .m2v
Rog

I'm pretty sure you can do that already (with a little editing of your AVS file of course :)) Or, perhaps Nic/dragongodz could add an option similar to CCE's - disable audio - in the 'Audio Codec' drop down box?

//Edit
Also, just wondering, what are the chances of an internal resizer? This is mainly for those whom have no knowledge of AviSynth and just want to load the AVI file - which doesn't have a compliant SVCD/DVD resolution - directly into QuEnc. Perhaps you could allow the user to choose the output resolution? And, of course an option to disable the option for those that already know how to resize in AviSynth :)

Peter Cheat

8th November 2004, 02:22

I wouldn't bother with internal resizing. AviSynth has far better resizing than ffmpeg. Plus Nic would need toallow AVI input, which wouldn't be hard to implement but would cause more problems (see what problems have popped up when muxxing was added). AVS->M2V is the basic idea of the app I believe.

I have a request too :D
More stats like peak bitrate, maximum bitrate, average quantiser, minimum and maximum quantiser etc etc. I've written the code for all of this in NuEnc, just haven't finished arranging the GUI (I hate GUI programming).

burnman99

8th November 2004, 07:24

Originally posted by Paced
I'm pretty sure you can do that already (with a little editing of your AVS file of course :)) Or, perhaps Nic/dragongodz could add an option similar to CCE's - disable audio - in the 'Audio Codec' drop down box?

Well i Can do video only buy using the KillAudio command but is there a way to do AUDIO only?! This would help!

Thanks!

Rog

Teegedeck

8th November 2004, 09:35

An idea for better scaling for high-bitrate matrices in 2-pass: If the summ of values in a custom-matrix is below, say, 1300 a high-bitrate matrix is recognized and the first pass performed at quant=3 instead of quant=2. I would very much like such a thing in XviD, too, BTW...

Paced

8th November 2004, 12:13

Originally posted by Peter Cheat
I wouldn't bother with internal resizing. AviSynth has far better resizing than ffmpeg. Plus Nic would need toallow AVI input, which wouldn't be hard to implement but would cause more problems (see what problems have popped up when muxxing was added). AVS->M2V is the basic idea of the app I believe.

Sorry, I wasn't more clear. What I meant was in its current state, whenever an AVI is loaded directly into QuEnc, I think an AVS file is created for it. I'm not sure if this is possible, since I am but a mere peasant in software development :), but perhaps Nic could somehow tell QuEnc to analyse the AVS input to determine its original resolution, then apply the appropriate calculations for correct resizing (4:3, 4:3 letterbox, 16:9, etc), then add it to the script? Probably more trouble than it's worth, even if it were possible...

Paced

8th November 2004, 12:14

Originally posted by burnman99
Well i Can do video only buy using the KillAudio command but is there a way to do AUDIO only?! This would help!

Thanks!

Rog

You can load a .wav file directly into QuEnc, then it will encode it to MP2/AC3.

Inc

8th November 2004, 15:30

Originally posted by Paced
Sorry, I wasn't more clear. What I meant was in its current state, whenever an AVI is loaded directly into QuEnc, I think an AVS file is created for it. I'm not sure if this is possible, since I am but a mere peasant in software development :), but perhaps Nic could somehow tell QuEnc to analyse the AVS input to determine its original resolution, then apply the appropriate calculations for correct resizing (4:3, 4:3 letterbox, 16:9, etc), then add it to the script? Probably more trouble than it's worth, even if it were possible...

That would mean an integration of a FitCD like applikation "within" Qenc. As FitCD does use the avi-input where a avs script will be generated incl. all PAR based resizings.

Resizings are too complex as they could be integrated in a such newbie friendly way without that advanced users will ran up the walls cause of too less teakable functions ;-)

So FitCD as speparate appl. should be your first choice as you can determine everything whats needed.

But if some c++ resizing routines are needed, Nic could look into Gripfits sources ... but as said above, 90% of the users will do the separate FitCD way.

bergi

9th November 2004, 12:21

Don't forget my request for separated passes in 2pass mode. :)

SILICON

9th November 2004, 17:32

The 3pass or 4pass mode.
For a best quality :-), and more encode time :-(

Mug Funky

11th November 2004, 01:14

i second whoever it was' request for 2-pass fit-to-filesize when the number in the bitrate box is over 15000 (that's the mpeg-2 max isn't it?).

this is probably very much a libavcodec issue, but is nonlinear quantscale possible at all? i'm not sure if it really makes a big difference in the end...

apart from that and what's been said here already, there's not really anything more i'd want out of quenc.

Shubin

11th November 2004, 15:37

A possibility to recover or to continue encoding.
It's for people with old computers.
There are possibly some program states (complete frames I guess)
that could we stored on the HDD and in case of a crush or manual
interrupt the encoder will be able to continue from that stored
state.

I'm not sure that many people would vote for this, but...
just an idea.

Another idea is to be able to store all options in a config file.

Mug Funky

12th November 2004, 06:03

Shubin: i second that one heartily.

i keep getting crashes at about the 35000 frame mark (haven't tracked the problem to anything specific yet, maybe it's a b0rk mpeg2 stream i'm reading from?), and a resume option that maybe continued from the last i-frame (for both 1st and 2nd pass) would be awesome.

a feature like that would really set it apart from other encoders, too. it'd save me a lot of time as well (imagine a batch file that simply resumes constantly until the encode is finished - i'd be able to do overnight encodes and actually have them finished the next day without that disheartening "access violation" error).

[Admiral]

15th November 2004, 08:51

I would also appreciate an internal resizer, though because there isnt one I've been forced to bother learning Avisynth finally lol.
though i disagree that it would make it less user-friendly..
@Peter Cheat: it already DOES support avi-input with this release..

but uh yea.. suggestions are more really just problems that I have....
when I pause the encoding and then resume it then misrepresents the current FPS... like if i was encoding @ 11fps before hand and then I pause for 30minutes and then continue it'll suddenly say its going @ 1.8fps.. and because of that the 'remaining time' will be HUGE..
sometimes it 'catches up' and finishes when i thought it would have, but other times it actually DOES take a long time to finish..
this problem also applies when I start running something else in the background that occasionally takes up some cpu usage (i.e. newsgroup decoding etc). though I have QuEnc's priority set to Low, it never seems to 'catch up' once the cpu has been relieved of all other requests... bleh. yea. but anyways...

my other suggestion is a built-in bitrate calculator, or even better, a system where you just punch in the final size you want, the bitrate of the audio, and it just does it itself (sorta like in gordian knot)

oh and one last suggestion - batch encoding?
I'm trying to re-encode over 100 ~20min divx/xvid/mp4 into dvd-r so yea that would be much appreciated! otherwise i may have to go back to tmpegenc :-( (SLOW!!)

okay yea sorry to sound like a big noob. lol. but yea I am when it comes to mpeg-2

Mug Funky

15th November 2004, 11:01

you can do batch encoding just fine from the command line... just make a .bat file

of course, you'll still have to make the AVS files, but i find it quite easy to just have videos with consecutive names and go through with a text editor. you can use regular expressions if you want to be tricky.

[edit]

oh, just thought of something else: perhaps a way to set b-frame ratio and offset, like in xvid? would this clutter up the GUI too much? (i don't actually know what quenc does as far as offset and ratio).

Peter Cheat

16th November 2004, 06:29

I don't think Nic wants the ratio and offset to be played with. It has a dramatic effect on quality (usually negatively). He has it set to some good defaults. Allowing the user to change is would cause many problems I think.

Bugsy

18th November 2004, 14:13

Originally posted by burnman99
The ability to enocde audio & video separately & save as .mp2 & .m2v
A template for VCD/SVCD/DVD.

That's about it. This program is Great!

Rog

Yes please! Encode to separate audio and video files. A free authoring program like DVDAuthorgui requires separate audio and video. I know I could fiddle the .AVS to give me video only and another tool to give me audio, but why not give the user an option to choose between elementary streams and a muxed file?

Bugsy

burnman99

19th November 2004, 20:21

Originally posted by Bugsy
Yes please! Encode to separate audio and video files. A free authoring program like DVDAuthorgui requires separate audio and video. I know I could fiddle the .AVS to give me video only and another tool to give me audio, but why not give the user an option to choose between elementary streams and a muxed file?

Bugsy

Bugsy, here is a workaround until it is updated...

For the .m2v file:

DirectShowSource("C:\Movie.avi", audio=false)
ConvertToYV12()

For the .mp2 or .ac3 file:

DirectShowSource("C:\Movie.avi", video=false)
ConvertToYV12()

This should give you both the .m2v & .mp2/.ac3 files. You can thank
clickit on the kvcd forum for the scripts.

Hope this helps.

Later!

Rog

Bugsy

23rd November 2004, 14:11

Originally posted by burnman99
Bugsy, here is a workaround until it is updated...

For the .m2v file:

DirectShowSource("C:\Movie.avi", audio=false)
ConvertToYV12()

For the .mp2 or .ac3 file:

DirectShowSource("C:\Movie.avi", video=false)
ConvertToYV12()

This should give you both the .m2v & .mp2/.ac3 files. You can thank
clickit on the kvcd forum for the scripts.

Hope this helps.

Later!

Rog

Thanx Rog!

Wanted to thank clickit on the kvcd forum, but they wanted me to pay to register....? Bad idea! I'm sticking to Doom9.

Bugsy :devil:

PS! Any idea of when to expect v 0.57 or better 0.60 or even better: 1.0?

Nic

24th November 2004, 10:49

PS! Any idea of when to expect v 0.57 or better 0.60 or even better: 1.0?
Maybe soon, I was waiting for the new DVD patch to be submitted to CVS. And it finally has....

-Nic

Koepi

24th November 2004, 11:01

Very nice, can't wait to test it :)

Keep it going Nic!

Regards
Koepi

Nic

24th November 2004, 11:18

Keep it going Nic!
It's hard to....Half-Life 2 is quite addictive :)

-Nic

Koepi

24th November 2004, 14:01

Damn, I have a machine at home which is fast and powerful enough for gaming in the meantime (Barton 2.8k, Radeon9.6k, 1GB DDR-333 dual channel memory on msi k7n2 delta-ilsr, 19" TFT /w DVI,...), but I lack the time for any games besides windows (will it work? will it stay (virus-)clean? ;) ) or "let's see if it helps functionality if I move this code block there..." - and of course "let's see what we can stress this new QuEnc build with" :)

So, come on, take the few minutes so we can play too! *hide*

Cheers
Koepi

Mug Funky

24th November 2004, 17:29

koepi: stop bugging the devs... you gotta give a little back to the community as well, you know :D

*hiding too*

[edit]

hehe... btw, VHQ for b-frames rocks. thanks koepi :)

Koepi

24th November 2004, 17:44

Originally posted by Mug Funky
koepi: stop bugging the devs... you gotta give a little back to the community as well, you know :D

*hiding too*

[edit]

hehe... btw, VHQ for b-frames rocks. thanks koepi :)

But but but.... that sysKin's code! :eek:

Enough of the off-topic. Let's patiently wait for Nic frag'ing away all his enemies :)

Cheers

Koepi (yay, xmas market opens tonight, I'll have some "hot wine" in some minutes!)

Teegedeck

24th November 2004, 22:38

Hey, I prefer 'Feuerzangenbowle' these days! :D

But that hot, spicy wine is also yummie!
...an' cheers to you!

Errr... Excuse me being offtopic once again *trying to strike himself*

Nic

25th November 2004, 13:50

(Got even more side tracked...got the SNOW codec compiling in Visual Studio, so I can use all the normal tools I use to optimise stuff....But a new release will be soon-ish)

dragongodz

26th November 2004, 01:20

:sly:

zilog jones

1st December 2004, 17:15

I'd also like to see an integrated bitrate calculator. I can't seem to find any free DVD video calculators anywhere, outside of going half-way through the project wizard in TMPGEnc, or using a *real* calculator (they're so passe!).

Bogalvator

1st December 2004, 20:23

http://www.videohelp.com/calc.htm

Works well for me......

freelock7

2nd December 2004, 08:08

a <shutdown > after encoding would be nice (like EclCCE feature).

dragongodz

2nd December 2004, 12:05

a <shutdown > after encoding would be nice (like EclCCE feature).
QuEnc 0.56 already has it for CLI use so i assume you mean add it to the GUI ?

freelock7

2nd December 2004, 18:02

Yes.

spase

3rd December 2004, 06:02

Originally posted by zilog jones
I'd also like to see an integrated bitrate calculator. I can't seem to find any free DVD video calculators anywhere, outside of going half-way through the project wizard in TMPGEnc, or using a *real* calculator (they're so passe!).

i found a free one called PowerBIT.

it appears to be in german or something, but you can right click the interface to switch languages. i dont remember the website, but i am sure you can google it.

spase

3rd December 2004, 06:08

hmm it has come to my attention that minimum bitrate would be a useful option as well (maybe only to me so that it doesn't studder on my standlone... but i can't be the ONLY one with a standalone that doesn't deal with low low bitrates).

oh yea and 3 or 4 pass would be nice as well... these are the only 2 real reasons i have been using NuEnc.... man i have 'wasted' a lot of time with that program trying different settings, matices, etc.

dragongodz

3rd December 2004, 12:28

hmm it has come to my attention that minimum bitrate would be a useful option as well (maybe only to me so that it doesn't studder on my standlone... but i can't be the ONLY one with a standalone that doesn't deal with low low bitrates).
the reason QuEnc doesnt have it is because with the current avcodec if you use anything other than a min of 0 the rate control goes nuts. thats one thing Peter has fixed in his rate control. :)

spase

3rd December 2004, 13:35

i see. i thought this might be the case.

Guest

6th December 2004, 14:22

Hi,
I'm trying to to get QuEnc to render out a mp2 file - but my avsvideo is generated in avisynth and not loaded(I have to use the images for rendering a fadeout2 on the sound), so I can't use the:
DirectShowSource("C:\Movie.avi", video=false)
Any other idears of how to kill the image in avisynth?

Btw. anyone knows anything of the how the muxxing is comming along in QuEnc(so I don't need this (m2v+mp2+mplex commandline)workaround?

Tin2tin

My DVDslideshowGUI (http://wxbasic.sourceforge.net/phpBB2/viewtopic.php?t=668) using avisynth and QuEnc.

dragongodz

7th December 2004, 13:02

anyone knows anything of the how the muxxing is comming along in QuEnc
well some dvd muxxing patches have been commited to the FFMpeg cvs so muxxing in the next version should be better. will just have to wait and see. :)

unixfs

7th December 2004, 16:41

It would be very nice if avcodec/Quenc encoded dark (and in general uniform) scenes without blocks
(on my TV I see them everywhere, whatever the bitrate, even with qscale=2).

Another artifact that avcodec always produces is ringing:
have you ever noticed how badly are encoded lights
cast by lightbulbs?

I understand these are a major issues (common to xvid/divx), but
you asked requests :)

Boulder

7th December 2004, 17:22

Have you tried using Blockbuster? It can work wonders on dark areas.

Which matrix did you use?

unixfs

7th December 2004, 17:25

I ran a short test with blockbuster, but the source was clean, and
the result was even worse than the standard one.
I never use custom matrices.

Boulder

7th December 2004, 17:53

Originally posted by unixfs
I ran a short test with blockbuster, but the source was clean, and
the result was even worse than the standard one.

And exactly with what settings? Blockbuster doesn't remove existing blocks, it prevents them from appearing in the low-lit areas in the encoding process.

I never use custom matrices.

Maybe it's time to try then? ;)

unixfs

7th December 2004, 20:13

it was a month ago, now I don't remember what settings I used for
blockbuster. I will retry later; what custom matrices do you suggest
for a 720x576 at 2000 kb/s ?

Boulder

7th December 2004, 20:32

Try QLB or Notch. Notch was developed mainly with the low-lit areas in mind so it might help reducing DCT blocks.

Try Blockbuster(method="noise",variance=0.5,seed=1).

unixfs

8th December 2004, 12:57

I just tried your advices: I can still see those blocks, both when
I use QLB alone and when I combine it with blockbuster.
The other settings are: 2passes, mpeg2, 2000 kb/s, dc 11, gop 15,
no bframes, high quality.

Boulder

8th December 2004, 13:03

Could you post a screenshot of a problem frame, preferably both the original and encoded? I'd be interested in seeing the actual frame.

dragongodz

8th December 2004, 13:36

settings are: 2passes, mpeg2, 2000 kb/s, dc 11, gop 15,
GAK!!! BAD, BAD and plain BAD.

1. dc 11 is for high def(high level) footage only, meaning realy big bitrate encoding(15Mb/s or more type encoding). dvd for example can only use 8,9 or 10. for lower bitrate encoding you may want to try a lower value such as 8 or 9.
2. no B frames with dvd frame size but a bitrate of 2000 ? use B frames(i reccomended 2 for dvd), its really is a must for so little bitrate.

unixfs

9th December 2004, 08:22

tested with 8/9 dc and 2 bframes.
Even so I can't see any difference

Inc

9th December 2004, 10:03

Do postprocess your "source" before, as if already minimal dct artifacts are given by the source, these will increase significantly when reencoding.

dragongodz

9th December 2004, 13:44

Even so I can't see any difference
none at all ? that is strange.

Do postprocess your "source" before, as if already minimal dct artifacts are given by the source, these will increase significantly when reencoding.
yep, try some denoising filters such as undot and/or fluxsmooth and see if they make any difference aswell. but please realise that you are using a fairly low bitrate for a dvd size picture, its not going to be the same quality as the original.

kirby55

9th December 2004, 14:21

I have used QuEnc for encoding a movie with a bitrate of 2100(JUST : DCT 10 - 2 passes vbr - scene detection ) and the result is very very good (the next time i put dct 8 when i have low bitrate)

my source was a dvd and this my script :

LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\MPEG2Dec3.dll")
MPEG2Source("D:\*****\******.d2v",cpu=2,idct=2)
Crop(26,0,-26,-0)
Undot()
BicubicResize(668,576)
Fluxsmooth(3,3)
Asharp(1,4,0.25)
AddBorders(18,0,18,0)

david

and sorry for my english, but me, i understand me :D

dragongodz

9th December 2004, 16:59

and sorry for my english, but me, i understand me
your english is fine so no need to apologise. :)

Boulder

9th December 2004, 17:11

Originally posted by kirby55
I have used QuEnc for encoding a movie with a bitrate of 2100(JUST : DCT 10 - 2 passes vbr - scene detection ) and the result is very very good (the next time i put dct 8 when i have low bitrate)

my source was a dvd and this my script :

LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\MPEG2Dec3.dll")
MPEG2Source("D:\*****\******.d2v",cpu=2,idct=2)
Crop(26,0,-26,-0)
Undot()
BicubicResize(668,576)
Fluxsmooth(3,3)
Asharp(1,4,0.25)
AddBorders(18,0,18,0)

david

and sorry for my english, but me, i understand me :D

You would probably get even better quality if you used 16-pixel borders. The encoder might not like the extra 2 pixels as they do not fill a complete block/macroblock.

kirby55

9th December 2004, 23:10

Add job like CCE ;)

kirby55

9th December 2004, 23:13

your english is fine so no need to apologise. :)
:D

When the next version ?

dragongodz

10th December 2004, 01:07

when Nic is ready. 1 thing i have found from working with Nic on programs such as Rejig and QuEnc is not to mess with it while he is messing with it....or when its time to intergrate it all its a mess. :)

freelock7

10th December 2004, 08:29

MPEG2Source("D:\*****\******.d2v",cpu=2,idct=2)

For a bad source or noisy picture, it is better to use cpu=4.

Bugsy

17th December 2004, 11:17

In my Christmas stocking I would like some templates for QuEnc.
The source is different NTCS/PAL, framerate is different, Interlaced/non interlaced, DV-PAL (AVI-1/AVI-2), noisy/clean source, AVI form VHS captures...... And output is different, ie bitrate 2000 kbps og bitrate 8000 kbps , GOP structure, interlaced/non interlaced/progressive, MPEG-1/MPEG-2, and much more....
Encoding speed, AviSynth settings (filtering/cropping/resizing)...

I would like some templates (and AVS scripts) setting the encoder up according to source and demand/preferences. Either a quick, low bitrate output or a slow high quality high bitrate DVD compliant output... I get frustrated reading a post discussing GOP structure, I/B/P frames and matrices just in the end finding out that the source is XviD....

Please....?

Bugsy

I have a feeling this came out rather negative... I'm really not... I love the tool and the discussions.

;)

PatchWorKs

23rd December 2004, 10:58

Support more codecs ! :D

(not just mpeg-1 & 2, but Theora, Snow, Dirac, XviD...)

lithoc

10th January 2005, 23:17

I'm trying to encode a VOB over the network.

VOB > network > dvd2avi > avisynth > QuEnc

I always notice that QuEnc only use ~90% of CPU usage.

Can QuEnc increase the input buffer so that it will buffer few frames advance before it actually process?

Thanks

BTW, is there anyone know howto:

VOB > dvd2avi > avisynth > network > QuEnc

Then I can utilise both PC resource :D

PatchWorKs

11th January 2005, 10:01

As always: FairUse' style version (direct DVD encoding)

This would mean a perfect backup application even for newbies, so it can be boundled with standalone players (and the author could obtain a revenue for his work...)

FU source code (http://www.fairusewizard.com/Release/fu-src.zip)

DarkFoon

22nd August 2005, 05:45

I don't know if this is even possible (or beneficial) with MPEG1/2 but I'd like to see a 2-pass CBR in Quenc.

My reasoning is this: as I understand it, currently avcodec (and therefore quenc) has a problem with bitrate spikes that make DVD's glitch in certain DVD players. A CBR mode wouldn't do that, but CBR is vastly inferior quality-wise to VBR except in a few applications. But a 2-pass CBR would allow the codec to do its best, with what it has, where it is. (another thought just occured to me, is there a bit reservoir in mpeg2 encoding like there is in mp3?)

Also, I encode my stills (for menus) as CBR, but sometimes, I have to use really low bitrates that crapify the menus :-(

Anyways, that's my request.

Inc

22nd August 2005, 09:06

I'm trying to encode a VOB over the network.

VOB > network > dvd2avi > avisynth > QuEnc

I always notice that QuEnc only use ~90% of CPU usage.

Can QuEnc increase the input buffer so that it will buffer few frames advance before it actually process?
I dont know your networks bandwith but it seems the resulted d2v Project needs too much time to get the real VOBs Data.

VOB > dvd2avi > avisynth > network > QuEnc

Then I can utilise both PC resource :D
?

If you open an avs script using an encoding or displaying software, then the avisynth.dll on the ACTUAL machine is used as inteface. Means with the graph you did above, still the main CPU work will be done on the Encoding machine, there's just the HD and Network Transfer of the Vob Host PC

dragongodz

22nd August 2005, 12:19

I don't know if this is even possible (or beneficial) with MPEG1/2 but I'd like to see a 2-pass CBR in Quenc.
its been able to for a long time. just untick VBR and tick 2pass and there it goes. at the moment the CBR made is actually crap(doesnt do CBR but a more constrained VBR) and i would not suggest really using it if you really need 100% CBR. and no, 2 passes shouldnt really be needed for CBR.

as I understand it, currently avcodec (and therefore quenc) has a problem with bitrate spikes that make DVD's glitch in certain DVD players.
yes but the latest QuEnc is not using libavcodecs rate control in an attempt to fix that. always a good idea to set the max bitrate a little bit under anyway.

Also, I encode my stills (for menus) as CBR, but sometimes, I have to use really low bitrates that crapify the menus
if you are encoding a still you could try constant quant at 2 or 3 etc.

DarkFoon

22nd August 2005, 17:11

its been able to for a long time. just untick VBR and tick 2pass and there it goes. at the moment the CBR made is actually crap(doesnt do CBR but a more constrained VBR) and i would not suggest really using it if you really need 100% CBR. and no, 2 passes shouldnt really be needed for CBR.
I really got the idea from WMV9 about 2-pass CBR, and there it increases the quality. So you are saying 2-pass cbr on mpeg2 doesn't really increase the quality?

DarkFoon

25th August 2005, 00:47

I'm back, with another feature request.
Perhaps a mode where it re-encodes GOPs with out-of-spec bitrates or ones that are outside of user defined bitrate.
I know of a few programs that do something similar (but I have to pay for those) And it seems like a simple implementation, except for needing to add a bitrate scanner or something into it.
thanks, bye

Mug Funky

26th August 2005, 08:35

i'm not sure, but i think cuttermaran can do it if it's set on "DVD compliancy mode" or something like that. i've not used it though.

it'd definitely be cool to have an easy tool to enable segment re-encodes. the only tool i know of is DVDmaestro, and that's not exactly cheap (or even very available). this isn't a quenc thing of course.

ooh! - manually set i-frames would be really REALLY cool. this would allow chapter points (without having to rely on scenechange), and also with some messing around it'd allow multiangle encoding where the 2nd angle uses the same GOP structure as the first (without having to resort to constant GOP structure). maybe an option to input a d2v file and use the frame type decisions in that (including RFF flags?).

okay, this is just becoming a personal wishlist... i don't really expect this stuff to be implemented into QuEnc :)

Sagittaire

26th August 2005, 11:32

QuEnc use Libavcodec binaries ... perhabs good idea to make very advanced codec setting:

- MP@ML and MP@HL profil
- Advanced ME setting (ME fonction, EPZS diamond size ...)
- Adaptative quantisation (Lumi, Dark, Temporal and Spacial Masking)

shirohamada

28th August 2005, 08:55

a confirmation to overwrite same name file

yes, i made a very dumb mistake :o

IgorC

28th August 2005, 19:17

It has already mentioned but multipass (3d pass as in Libavcodec)

Inc

28th August 2005, 23:24

according to ...

a) multipass: PeterCheat did that mod to libavcodec in his NuEnc AFAIK but Im not shure if its worth the sgnificant more time it needs to do the whole job.

b) Blocks on low lit scenes as mentioned a page before: I clearly remember the time (1/2 year ago) when we did test mencoder in its settings and no matter what happened, blocks on such low scenes where obligatory. Now Libavcodec uses a linear Quantizationscale and a Quantiz.Factor of 2 is not that less in linear Q encodings (Im not talking about FIXED Quantscale).
I only fixed that using Vmin=1 and very low lmin settings as it did lower the low Q barrier from 2 to 1. The Encodings on these Parts then resulted perfect. Petercheat advised me NOT to use "Vmin=1 in combination with trellis" as then the rumble could start. I never used trellis so I went happy with the parameters above.
Im still using Mencoder for some testings and since 2 Month I got not only one bitrate spike on total different Movies. It seems that something really has changed since 1/2 year? Im not shure or the luck was with me the last 2 Month.

Zarxrax

2nd September 2005, 03:50

I just tried QuEnc for the first time tonight. Actually I have been trying a lot of mpeg2 encoders today--CCE, HC, tmpg...

QuEnc made my output file at half the size of the other encoders. Looks like theres some serious issues with the rate controll there ;p
But despite the small file size, the output file completely KILLED all the other encoders in video quality. I couldn't believe it. All of the encodes from the other encoders honestly looked like shit on this video, and yet here was this QuEnc encode at HALF the size of the rest of them, and it looked acceptable. Source was anime/cg material.
Anyway, that was my experience with QuEnc tonight. :)

shirohamada

2nd September 2005, 11:49

try switching off trellis and/or extreme & slow.

has anyone requested for stats file output ?

Inc

2nd September 2005, 13:55

QuEnc made my output file at half the size of the other encoders.

Now .. the question: How is that anime in its treatment? I mean many low lit scenes?
Just an assumption of mine: Qenc uses Vqmin=2(?) and on such LowLit/LowComplex scenes it won't lower than Q=2 even it would be useful on LowLit/LowComplex scenes. By keeping the Q at min=2 this (in this case unfortunately) safes bits but keeps the Q min = 2.

If you want to "tame" avcodec's mpeg2 encodings there is a nice way:

Do include in your avs script a KeepMin() and a KeepMax() Function.
Now, such a function KeepMin() would include a very well performed adaptive usage of bockbuster(), means a fine spray of noise would be used ONLY on these mega low lit scenes. By this the avcodec rises its bitrate, Q lowers and the nice sideeffect is also the known advantage of a little spry of noise on such plain low lit scenes.
The KeepMax() function is a bit more complex as it has to recognise the Bitstream in its complexitivity based on a Luma Diff. to the prev or next frame incl. maybe a good edgemask. Now, IF that complexitivity-detection does see something like a big uneasy frame/part like in "Poltergeist"'s TV noise on the zoom to the TV screen, then it will just "break" the complexitivity using a tiny adaptive blur. On my tryouts the output was MUCH better then it would have been too much quantized and the Spikes did fall down from 11000 to 8000kbits.

But the bad sideeffect of all this is, that such 2 functions will use their needed CPU power, so by this expect an encoding speed decrease by about 200-300%

DarkFoon

2nd September 2005, 17:38

@incredible
have you made such functions? If so, please do post them!
(Or am I an idiot, and they've already been placed somewhere really easy, like in the AVS docs?)

Zarxrax

2nd September 2005, 23:31

Just an assumption of mine: Qenc uses Vqmin=2(?) and on such LowLit/LowComplex scenes it won't lower than Q=2 even it would be useful on LowLit/LowComplex scenes. By keeping the Q at min=2 this (in this case unfortunately) safes bits but keeps the Q min = 2.

You seem to be correct. I set it to encode at constant quantizer 1, and the file came out twice as large. Could an option be added to QuEnc to allow it to use quant 1? Also how about a queue to encode batches of files...

I found out its the "high quality" switch that is really doing wonders for the quality. On other encoders I would get lots of "smearing", but that isn't present on QuEnc if I use this switch. Unfortunately though, QuEnc DOES throw in tiny gray blocks all over the place. I wonder if this is due to any particular option, or if its just a bug in ffmpeg?

Correction: not tiny gray blocks, but maybe luma inverted blocks, or blocks jumping around places they shouldnt be, or something... hard to describe.

Inc

2nd September 2005, 23:58

Im not talking about Quantscale=1 ! Im talking about the VARIABLE Quant curve to be able to go below 2 means min=1 ant Vqmin.
Thats a very nice behaviour I noticed on libavcodec that he often behaves in its Q curve in a very good relation to the Bitrate.
Thats how it should work, if the bitrate drops significantly it rises the Q.

Shure a Quantscale of 1 does result (also) in a very inefficient blow'n up target ;)

Zarxrax

3rd September 2005, 00:55

Ok, I suppose the quant 1 doesnt really make a whole lot of difference in my case, as I still get these bad artifacts in any event. Here's a sample of them:
http://amvhell.com/blocks.mpg

You should see it around the edges of the letters. Like I said, the High Quality switch really helps make it look better. On other encoders like CCE, it just starts smearing all around outside the letters.

DarkFoon

3rd September 2005, 00:59

@Zarxrax
that wierd luma block jumping is a very well-known bug in libavcodec trellis with quant = 1 (I think).
turn off extreme&slow and trellis and it should go away (you can leave high quality on, because it doesn't use trellis, if memory serves me)
I thought that was fixed a while ago, though...

Zarxrax

3rd September 2005, 02:49

I did not have trellis or extreme enabled when I encoded this clip. Note that this is a completely different kind of artifact from the white blocks that appear when those options are turned on.

Mug Funky

3rd September 2005, 07:49

an idea occurs as far as 2-pass CBR goes - on the first pass it can determine compressibility (obviously) and assign quant matrices per-GOP for the second pass.

it could use the default matrix (be it QLB or default MPEG or whatever), and change the matrix when either (a) it's at q=1 and the bitrate is below the max by a certain percentage, or (b) the bitrate exceeds max.

if the generated matrices are simply scaled (and rounded up or down) versions of the input matrix, then it wont throw rate-control off all that much - all that would affect it is rounding error in the matrix values.

so long as it keeps matrix values above 8 (for compatibility), this should use as many bits as possible without resorting to padding with zeros as much. it should also be DVD compliant.

it'd be sort of like a simple version of autoQ that doesn't strive to get perfect details, but merely give finer control over the quantiser (q=2 with an mpeg matrix where all values are halved is the same as mpeg matrix at q=1, not accounting for rounding).

DarkFoon

3rd September 2005, 09:04

@ mug funky
Idea seconded!

@Zarxrax
hmmm. I have watched your clip, and I see no artifacting at all. no random blocks of different luminosity. nothing.
I watched the edges of the text, and still I saw nothing. Perhaps its your mpeg decoder doing post-processing while decoding and its messing with the image?
When I play it back, it looks pristine.
waht are your encoding parameters?

Inc

3rd September 2005, 11:09

I also can duplicate DarkFoon's statement.
The encoding of the beginning is good. The part after the Noisescreen means the following anime is a disaster due to a jump of the q curve. Seems the internal Ratecontrol gets totally irritated when encountering that noisescreen part.
I always do test my mpeg encodings in VirtualDubMod and no directdraw accel. means native output no other filters do affect.

BTW. viewing in Bitrateviewer the encoding did not use constQuant!

@ DarkFoon
Ough, .... I tested a lot libavcodec based frontends but since a 1/2 year I havn't used/testet them anymore. I have to look at my disk or at the KVCD Forum where I maybe posted these functions. Ill have a look.

Inc

3rd September 2005, 11:20

@ DarkFoon
Here a copy of a posting of mine of Aug.2004 at KVCD forum (so its oooold ;) ):
In mencoder usage a got a line where a minrate setting did result in a very nice and smooth Q curve ... but only in 70% cases of encodings, so this libavcodec is very sensitive on what comes in!

I did make 2 functions to get libavcodec more under control, means ...

- one function which does add an amount of noise only at black scenes and keeping everything else untouched. This does keep the min bitrate at a level which you can tweak by thresholding.

- the other function is based on MA but more reduced, means: An unfilter will do blur heavy flickering/flushing scenes based on a threshold, so that filter gets only kicked in when such a scene is detected. And at such smashing scenes you wont recognise that threshold based kick ins (just IMHO).

Well, something to play with:

peakblur(int "amount of blurring", int "complexitivity threshold")

Peakblur () as Default means peakblur(2, 10)

Means at a defined luma (nf) "threshold" the blurring kicks in using an "amount" of blurring based on nf (like MA).

keepminrate(float "amount of noise" , int "blackness threshold")

keepminrate() as Default means keepminrate(1,50).
If you do rise the threshold even more dark scenes besides black frames will be noised. The rise of amount (floats are accepted!) speks for itself and should be used to get the wanted min bitrate which differs in case of resolutions.

You will recognise that the noise will be almost quantized off aferwards when previewing in Vdub using the Levels filter in Vdub to brighten the preview to max.

Copy these lines and safe them as textbased file using the .avsi suffix into the avisynth Plugins Folder. Next time in your script you can use these commands.

function keepminrate(clip c, float "amount", int "thresh") {
amount=default(amount,1)
thresh=default(thresh,50)
d=Blockbuster(c,method="noise",detail_min=1,detail_max=2,variance=amount,seed=8)
c=ConditionalFilter(c, d, c, "AverageLuma()", "lessthan", string(thresh))
return c
}
function peakblur(clip c, int "ml", int "th") {
ml= default(ml, 2)
th= default(th, 10)
c=ScriptClip(c, "nf = YDifferenceToNext()"+chr(13)+"nf >= "+string(th)+" ? unfilter( -(fmin(round(nf)*"+string(ml)+", 100)), -(fmin(round(nf)*"+string(ml)+", 100))) : last ")
return c
}
function fmin( int f1, int f2) {
return ( f1<f2 ) ? f1 : f2
}

You should place peakblur() BEFORE keepminrate()

Remember this is just a first go and shurely not a final state, so bugs/experiencies reports would be nice.

Inc.

The PeakBlur one should be really updated as masktools internal functions would do a much better job.

Zarxrax

3rd September 2005, 19:01

@Zarxrax
hmmm. I have watched your clip, and I see no artifacting at all. no random blocks of different luminosity. nothing.
I watched the edges of the text, and still I saw nothing. Perhaps its your mpeg decoder doing post-processing while decoding and its messing with the image?
When I play it back, it looks pristine.

Ok... you are right. I don't see these artifacts when I play it through virtualdubmod. It must be my decoder. The thing is, I could SWEAR that I've been seeing the artifacts in vdubmod too, prior to this encode... Maybe I'm just going crazy O.o