NuEnc 0.01b Released

[dragongodz]

In VBR-2Pass, at low bitrate, I did not see any real improvement in picture quality compared to the preceding versions even if the choice of the advanced options is much vaster and more pointed.

remember that the new rate control is trying to do exactly that, control the bitrate. the old rate control does not do a real good job. so the old RC can produce rather big spikes which will look great for those scenes but may not play happy on hardware. while the new RC should produce a more consistant bitrate which should produce a more consistant quality. sort of like xvids new stricter scaling option. thats the hope atleast.
as for all the extra options Peter has added to the gui....no comment.

[Peter1234]

Peter Cheat,
Is there a VBV buffer size problem? When I set VBV to 224 in NuEnc, BitRateViewer shows output VBV as 112.

[freelock7]

@Dragongodz
Yes.This is why the new version of QuEnc0.56 functions less better in the top bitrate.
My test are based on PC results not on TV. I know the underflow and VBV problems but I think that the goal of the new libavcodec-on the strict level of quality- is to succeed at the same result of the old version.It would be absurd to evolve and obtain worse performances, no?
But I was not satisfied with your conclusion.
So I tested once again in the morning and I found this strange result(I'm lost here):
QuEnc0.55Alpha -with the new modified libavcodec????-produces the same great quality as its little brother QuEnc0.54.
But-last but not least-QuEnc0.56 is worse!!!
What kind of libavcodec 0.55Alpha is using?

[Peter Cheat]

@Peter1234
I thought this too, but BV reports 1835008 / 1024 / 16 (not in KB) whereas NuEnc is asking for a size in KB (1835008 / 1024 / 8). Just there to confuse us .

@evade
I intend to upload all my sources, but I don't see a point in uploading everything while I don't have a reliable host. But I have found an awesome free host, just need to get things set up.

@freelock7
The bit distribution (and quality) depend on the rc equation used. Did you try increasing the value of rc variability to 1 (hint hint). QuEnc uses 0.5 by default, but because of the bitrate tolerance, its value is somewhere between 0.5-1 and varies throughout an encoding. Setting RC Variability to 1 is the same as CQ Mode (but not for 1-pass VBR). The bonus being that you know what file size to expect. The downside is you need to do at least 2 passes. RC variability changes the bitrate response - 0 means no change (CBR), 1 means change to keep constant quality (average quantiser).

@dragongodz
I added more options to clutter the GUI . Actually, I added the extra options because I was annoyed that I had to recompile every time I changed a parameter. The fact that it is more complicated to use doesn't hurt me .

I've done some testing for macro block based rate control (for potentially higher quality). The problem is that first pass stats file becomes MUCH larger than the actual MPEG. Might be better to just forget about that one. There is no real point for me to port the XviD rate control. I stole the good bits .

EDIT:
My aim was to fix the underflow issues. As a consequence, quality may improve as the bitrate follows a strict rule (the rc equation).

[Peter Cheat]

@freelock7
QuEnc0.56 may appear to be worse because Nic is using constant quant 2 for first pass, but I don't think he reduced the scene change threshold to compensate. You then get fewer scene changes detected (avcodec uses a stupid scene change algorithm which detects more scene changes with higher quantisers. Scene changes do not vary with quantiser to my eyes )

[dragongodz]

It would be absurd to evolve and obtain worse performances, no?

no. the aim is to stick to that standards and be hardware compatible. who cares how great it looks if when you go to play it on hardware it doesnt play or stutters like crazy. so yes everyone wants the quality to stay as good as possible but there will be times when it has to be reduced.

QuEnc0.55Alpha -with the new modified libavcodec????-produces the same great quality as its little brother QuEnc0.54.

that would be because 0.55 and 0.56 do not use the new rate control. Nic said he was planning 0.60 to be released in 2 versions, 1 with old RC and 1 with new RC.

I've done some testing for macro block based rate control (for potentially higher quality). The problem is that first pass stats file becomes MUCH larger than the actual MPEG. Might be better to just forget about that one

for 2 pass MB rate control is probably overkill. for 1 pass VBR however it would be very good.

My aim was to fix the underflow issues. As a consequence, quality may improve as the bitrate follows a strict rule (the rc equation).

or decrease in scenes that the old RC threw copious amounts of bits at.

[freelock7]

@Peter
VBR-2Pass comparison (2500kbs)
I confirm with Bitrateviewer: QuEnc0.55Alpha is better than 0.56 with a versatile response in high bitrate.
Peak in high motion picture at 8200 (5500 for QuEnc0.56).
[Settings:Scene detection/GOP15/DC10/QLB Matrix/interlaced]
@Dragongodz

no. the aim is to stick to that standards and be hardware compatible. who cares how great it looks if when you go to play it on hardware it doesnt play or stutters like crazy. so yes everyone wants the quality to stay as good as possible but there will be times when it has to be reduced.

I do not agree
I encoded with FreeEnc and it is read by my DVD player (some minnor underflow can appear sometimes).
I'm looking for the best picture quality on my TV screen. If not, I choose another encoder which works better.
So why CCE is so expensive?
Because it works fine with a good quality.

[Peter Cheat]

@freelock7
Set the RC variability to 1. The quality will be much better. Try it.

The idea is to maximise quality, and stick within the standards. NuEnc does both, but you have to chose RC variability to your own personal tastes. 0.5 introduces heavy curve compression, 1 is just linear even compression. FreeEnc uses 1 by default (rc_eq=tex).

So why CCE is so expensive?
Because it works fine with a good quality.

No, because they are rip-off merchants .

[Easy123]

@Peter Cheat:

If you´ve got a free minute could you look after the Link for NuEnc 0.01b on your Homepage. I´d like to give it a go, but the only thing I get is "Error 404"

[Peter Cheat]

Another mirror:
NuEnc0.01b

[johnnyquid]

Peter,

Since there are a lot more options in the GUI that might confuse people it might be a good idea to have a button that resets all the settings back to default settings.

[Mug Funky]

I do not agree
I encoded with FreeEnc and it is read by my DVD player (some minnor underflow can appear sometimes).
I'm looking for the best picture quality on my TV screen. If not, I choose another encoder which works better and everybody can tell you that.

not so. ask anyone who encodes for a living.

people don't worry about how good the quality is if the thing glitches out on their equipment. they'll only notice the glitches. that's a guarantee.

between a slightly blocky DVD and a slightly glitchy one, i can guarantee that the glitchy one will be returned more often (people feel like snobs if they complain about the quality... don't know why).

personally, i think standards compliance is THE ONLY thing stopping lavc based codecs from being used in the mainstream. once that hurdle is overcome, THEN we can make the quality even better. not the other way around, because you must bear in mind that the good quality we spend time trying to achieve might not be possible in a standard compliant model. we should work with the standards and put in as much quality as they'll allow, being careful not to break any possible compliance in the process.

otherwise we might as well author DVDs with VP6, or RV10, or any number of codecs that a DVD player will not understand.

it seems to me that the point of using MPEG-2 rather than a superior codec is precisely one of standard compliance - who wants good looking mpeg-2 if it doesn't play? we might as well be using a better codec.

[edit - minor grammatical adjustments]

[Peter1234]

Peter Cheat,
After you get all of the important things fixed, could you do something to more clearly indicate when encoding is complete. The progress bar color is not very different when starting and finished, and I usually end up looking at the Remaining Time number to confirm that it is finished. Maybe make progress bar change from gray to red. Ringing a bell when finished would also be a nice touch. I know you have already put a lot of work into this, hopefully you are willing to put in a little more. Thanks for all the work you have done already.

[freelock7]

@Mug Funky
It is explained much better but I never said anything else.
Quality & compliance are working together.
@Peter
No change with RC variability <1>.

[dragongodz]

It is explained much better but I never said anything else.

maybe english isnt your first language so you misunderstood but Mug Funky just said basically what i had already said. to which you replied "i do not agree". so actually you DID say something else.

[freelock7]

Sure! Here I'm talking french! So, excuse me for my bad explanations.

What you have to know, that's my research of the quality.
Everybody is trying to obtain such result by developing new product!
I believe in the great libavcodec potential.
That's all.

[Peter Cheat]

@johnnyquid
Good point. Probably why some are complaining about are quality issues. I'll throw that in next release.

@Peter1234
Another good point. I'll make it more obvious that encoding has finished too.

@freelock7
If there is no change when using RC variability = 1, then you are probably doing something wrong. Btw, did you try more than 2 passes? 3 passes gets to over 99% optimal. Also what settings are you using in FreeEnc that provides better results than NuEnc?

EDIT:
Perhaps I should add support for FreeEnc's templates and reduce the number of options in the GUI? Is this a better idea?

[dragongodz]

reduce the number of options in the GUI? Is this a better idea?

the eternal dilema, how much is too little and how much is too much.
its better IMHO to have people saying they wish an option was there than to have them complaining the output is bad because they frelled with things they dont understand(but think they do).

add support for FreeEnc's templates

what would be the point though ? to give people back access to things they shouldnt be playing with unless they really know what they are doing ?

[Peter Cheat]

Adding some sort of template support would be beneficial I guess because known settings could be put into files and users would select a template, bitrate, mpeg version etc and the output would be more predictable. Editting those templates would not be a good idea for newbies, but on the other hand, the only way to learn is to try, make a mistake, then figure out what you did wrong. For example, settings used for high quality in avcodec doesn't work well at low bitrates, especially trellis and RD MB decision. But I didn't know that until I tried.

Here is a mirror to all the files relating to avcodec. I haven't finished uploading all the sources, but I do intend to. I've rewritten the scene detection algorithm in avcodec, and now intend to implement closed gop with scene detection, the same way as done in CCE (only closing gops when there is scene change). Just need to figure out what the current code actually does in that area.

[dragongodz]

Adding some sort of template support would be beneficial I guess because known settings could be put into files and users would select a template, bitrate, mpeg version etc and the output would be more predictable.

yes i can see the benefit of templates but i just dont agree with the freeenc model of having every option possible, buts thats me.

now intend to implement closed gop with scene detection, the same way as done in CCE (only closing gops when there is scene change).

cool. another minor problem is open GOP streams do not start with a closed GOP even when B frames are used, so starts IP instead of IBBP(for example). so the first GOP is always shorter than specified. other encoders such as CCE and Tmpgenc close the first GOP in such cases from memory.