QuEnc
 

From doom9 forum, a mpeg encoder based on ffmpeg by Nic:

http://forum.doom9.org/showthread.php?s=&threadid=72827

Agrrr :evil:
You've outrun me by "turning computer on, and finding this on doom9" time :mrgreen:

Did you made some test :?: I will do some late evening today.

If it uses internal FFMPEG library, then we already know what the results will be ;)

-kwag

Think that answers your question, taken from http://forum.doom9.org/showthread.ph...540#post459540

Thanks GFR :)

I heard, and went running over to doom9 for more info.

I like it when new toys are discovered. Make things all the more interesting
sometimes. I am pretty busy with other matter at home, and try and find the
time to play :lol:

Cheers everyone,
-vhelp

Hi everybody!

I briefly tried QuEnc 0.2 with 352x288 PAL source at 700 kbps target.

It's pretty good for an early version (but not good enough for me for the moment). It uses a GOP size of 12, with IBBPBB... pattern.

The good:
-Is incredibly fast (comparing to ffvfw, I still haven't tried mencoder, I'm scared of what I see in that other thread, I believe I have to spend an entire week-end to begin to understand what is going on there :oops: ). I get about 50-55 fps with trellis and 60-75 without.
-I'm under the impression that the bitrate fluctuates much less than with ffvfw (the peaks are lower).
-The GUI is neat.

The bad:
-The GUI is too neat, it lets out very few options for the user. I hope this will change.
-The medium quantization is much higher than with ffvfw for the same source and the q line jumps like hell in BV. This gives easy noticiable quality fluctuation with the GOP. With 2 pass encoding (pretty useable since is very fast anyway) this jumpy q line is a little smoother but the quality is still badly fluctuating.

I hope things could change if we could have acces to more advanced options.

So for the the moment I stick with ffvfw (as it is in heavy developement, as I can see). But who knows... I'll wait for future versions anyway.
marcellus

@All,

Well, Nic has really done it now :)
I just tested his latest version of QueEnc ( 0.45 ), and the results are just excelent.
He has included KVCD's Notch matrix as an option.
Here's the latest version: http://nic.dnsalias.com/QuEnc.exe

Here's my sample with this version: http://www.kvcd.net/quenc-test.m2v

-kwag

Great! :D :D :D :D :D :D :D
Now I know what will be the encoder for DIKO... AVS support, full command-line, GOP and KVCD matrix... How could I ask for more?
I'll put QuEnc integration for DIKO this is high priority "To-do" list! Thanks Nic for the great work. :)

??

Full command line?

I did a test on a capture and the mencoder-by-hand way resulted in a better quantisation quality :?

:)

i didn't try doing quality tests yet. But mencoder can't accept AVS input in a automated way.

There is still some "timing" problems I've found. If you encode with open GOP, when playing back, the movie skips and jumps all over the place.
I'm currently testing with closed GOP, and so far, seems to work correctly.
There's a problem with the encoder when creating short GOPs. Looking at the GOP with an analyzer, I can clearly see the short structure messed up. Look:

Look at the structure when the GOP size is 3 :!:
There are no P frames.
So a decoder will some times see IBBIBB other times IBBP or IBBIBBPIBBPIBBIBBIBB
so the decoder will "think" to expect IBBP, when in reality, it will receive IBBIBB some of the times. This is throwing off the decoder's sync.
So for the time being, closing the GOP seems to fix this, because the GOP is larger.

-kwag

@vmesquita,

The good thing about this GUI/Encoder combo, is that we can forget about file size prediction :)
I already encoded Red Planet, targetting 1,332,485KB, and the final encode size was 1,365,477KB. So this is really sweet :D

-kwag

Yes, I encoded a clip from Shrek and it looks amazing! Only one problem, when I burned the clip and watched it on my stand alone DVD it looked almost interlaced?! Any ideas why? BTW I encoded the clip with open GOP. Will try another clip with closed GOP...

Was this PAL :?:
I've only tested on progressive matereial, and then I run PULLDOWN.EXE on the .m2v, and then multiplex with audio. Looks great on the standalone DVD player. :)

-kwag

No, it was the NTSC version of the movie. So, maybe I should try running pulldown on this sucker to make it work?

EDIT: BTW where can I download this program :?:

Definitely :D

-kwag

i only did tests changing to the KVCD notch matrix and 16:9,nothing more
i need to learn how to do/use command lines for QuENC!
who can help me?
some samples and hints are welcome!
:wink:






@ all:
hey, come back here..... :P

Where can I find this program :?: (PullDown.exe)

http://www.kvcd.net/PULLDOWN.EXE

Edit: Remember this is a command line utility :!:

Thanks. I found this - a GUI version called DoPulldown here:
http://www.doom9.org/Soft21/SupportU...wn.1.0.0.0.zip
Is it the same program exept for the GUI?

Yes, I believe so :)

-kwag

Burning right now...
I'll let you know in a second or two...

YES :!: That was it. Works like a charm :wink: Damn, this looks good. The quality of QuEnc 0.45 is really something else :!: :)
Thank you for the tip Kwag.

Now try encoding the complete movie using CalcuMatic (or other bitrate calcs) to get your average bitrate. Then see if the final size matches closely the wanted size ;)

-kwag

Kwag, the GOP 3 Frames issue to me seems like an Scene Change issue.

Yes it does. But only when very close scenes are detected. If you look when the GOP was 6, there is no problem. There should be a constraint in the encoder to limit the GOP size to a minimum of IBBP (4 frames), and that should correct the issue. Or really: minGOPSize = number_of_B_Frames + 2 ;)

-kwag

I think this issue can be resolved. ffvfw has an option for minimum interval between I frames and since is also libavcodec based I think that nic can expose that setting as an advanced option.

Anyway, QuEnc made a lot of progress but it's output is still under the quality of ffvfw (ffdshow actually) and the gain in speed does not pay the quality loss. But it's still 0.45 release...

Edit: Well, I forgot, another big advantage of QuEnc is VBV compliancy. I wonder why ffvfw is the only libavcodec incarnation thas has this issue, why milan does not fix it...

marcellus

As I got the actual QuEnc vers. .... where can you see a command line option??? My vers. does not show one, neither shurely to alter it as its not avaiable.

He should offer more options of ffmpeg in his GUI .... for example processing interlaced sources.

Man, it's very fast!!!!! :D 8)
I am currently encoding a AVI movie using the latest Dialhot v4 script:
With Trellis, GOP 18 and KVCD Notch I get 19.5 fps,
Without trellis, it goes up to 24 fps (realtime!)
Amazing speed.
Can't comment on quality yet, let's wait for the outcome. I am doing 2 pass VBR. I had previously encoded this movie using CCE so I'll be able to do a good comparison.

@incredible:
Do this:
QuEnc /?
A window will pop up with command-line options.

I have to retract what I said. In ffvfw the minimum I interval setting doesn't work as expected. With any value other than 1 for min I interval all GOPs are screwed, it puts a lot of consecutive P frames
Normal GOP: IBBPBBPBB....
Screwed (example): IPPPBBPPPPPPBBPPP..

Well, you can use VBVPatcher on your ffvfw encodes, and then the MPEG file will be "branded" compliant.
But the way I see it, I find QuEnc much more simpler and flexible than ffvww, and I'm getting extremely good results.
I think we will all be using QuEnc more often ;)

-kwag

@kwag
I agree with you, QuEnc is very good. I would like though to have all the libavcodec options out to play with them. :twisted: I hope it will be in 1.0 release.
I know that QuEnc has a very good quality right now but I'm targeted on small resolution (352x288) and any artifact is more annoying than with full resolution (I have double size blocks on TV screen). So thats why I am very picky and since the speed of ffvfw is acceptable at 352x288 (25-30 fps) and VBV issue is ignored by my player, you know what I choose for the moment. I am sure that QuEnc could do a better job with the right settings but we still have not access to them.
marcellus

As I posted earlier; my encode of Shrek looks amazing! Even though I encoded it in MPEG-2 and SVCD resolution (480x480). Have anyone tried QuEnc with MPEG-1 output? Results :?:

I did, and I also liked the MPEG-2 results better. Seems the encoder creates better results at MPEG-2.

-kwag

Yes. And another thing that's great about being able to encode good looking KSVCD's (MPEG-2) is that they're playable almost everywhere - on almost everybody's stand alone DVD players :D Even at my parents' house... :wink:

From what I understand it's best to use a closed GOP?
The setting I used for Shrek (NTSC) were the following:

Main window
* MPEG-2
* Use VBR - Checked
* High Quality - Checked
* Use Trellis Quant - Checked (But have no clue what this is :roll: )
Advanced options
* Force Closed GOP - Checked
* Use KVCD "Notch" Matrix - Checked
* GOP Size=18
* Max B-Frames=2
* 4:3 Aspect Ratio

Then, in the main window, I simply pasted the average bitrate I got from CalcuMatic and clicked Encode. Ta-da!
Please, inform me if I should use some other settings for my encodes.

Question: Can we use the GOP from the TMPEnc templates with QuEnc (PAL=25, NTSC=30, Film=24)?

@Kwag
Very nice quality on your Red Planet clip :!:

EDIT: If you use HybridFuPP to encode your movies - Take the average bitrate from CalcuMatic and multiply it by 1.05 :wink:

Well, my test got a very wierd result. I asked to encode 480x480 at 976 kbps, and used trellis, 2 pass VBR, Notch, GOP 18.
The file came out at 725 kbps. I opened it with BitrateViewer, and The Q line was constant at 2.67. I have the same file encoded with CCE at "real" 976 kbps, but still could not do a side by side comparisson. I opened the encoded file and it looked good, but in 17" monitor most stuff do. :D

You got them all right :D Yes, but I think this encoder doesn't like longer GOPs very much. Stick to 15(PAL) or 18(NTSC) Thanks :D No. It shouldn't make any difference, because the encoder is trying to maintain average bitrate.
BTW, what was your final file size on "Shrek", compared to the size wanted by CalcuMatic :?:

-kwag

kwag wroteActually, when I first encoded Shrek (my first try with QuEnc) with the average bitrate from CalcuMatic the final vide file got about 40 MB too small. But then I multiplied the avg.bitr. by 1.05 and the final movie - muxed - is 806 MB, perfect!

Note: When I use Incredible's Slicer routine to predict I also have to multiply the wanted sample size by 1.05 to get the final encode right.

EDIT: This is when I use HybridFuPP to encode the DVD's

Wait a minute....
So the average bitrate is not the average bitrate that the movie will be encoded? :? :? :?