Does GK Support DivX v5.1

[ThePanther]

Just A Quick One ...

Does GK Support DivX v5.1 ?

Thanks

[vorbis]

Its only released one day, give them some time to actually test it!

[ThePanther]

Sorry Guys, I was justing checking to see if anyone had ben using GK with the Beta, and therefore should possible work with v5.1.

[Doom9]

well.. why not try it out yourself? If they haven't changed the registry keys I could imagine that it works.. if they have, it most certainly won't work.
Don't you love the thrill of trying something out on your own? It is the factor that drove me to creating my own website.

[ThePanther]

I do, I am a developer myself, I'm just in work at the moment and haven't the time until this evening

[twist3d]

Hi,

it seems to work with gordian knot, i'm doing my first divx 5.1 encode (and first posting, i've been quiet reader of this forum for a couple of years).

The movie is "The Mission", with Robert De Niro and Jeremy Irons.

640x272, 1400mb is my target size.
Slowest mode, no gmc/psychovisuals enabled,
divx "home theater" compatible profile.

divx encoder settings:
-bvn1 1433.600 -psy 0 -key 300 -log "F:\mission.log" -b -sc 50 -pq 192 -profile 0

Encode box is winxp pro, athlon xp 1900, 512ddr.

- comptest without feedback window about 10fps
- credits encoding with quantizer 28 about 4-8fps ?! woo :)
- now encoding at steady 2-4fps

Let's see what happens :)

[jeremymacmull]

2-4 fps on slowest mode on 1900 XP that very very sloow

the real thing will be how much of a difference is slow / slowest mode in 2cds vs standard cos the encoding time is a huge jump not to mention phy on slow as well

ive got a p3 850 and at the mo using orignal 2 pass it take 7-9 hours per film i do which is fine for overnight stuff i get 9 fps roughly if i drop to like 1 fps its not worth while the extra quality as i am quite happy with 5.05 quality wise

JEREMY
ps dont say get a faster pc as i cant afford one

LOL

[kurdi]

Ok, seems to be significantly SLOWER than 5.05(?) --- wasn't it supposed to be faster or something! Also there seems to be some sort of problem muxing video and sound (I use Gknot 28.2beta btw).

[stax76]

nothing has changed except some new entries for new features, support would only break when they change the root key or the bitrate entry

[Acaila]

Ok, seems to be significantly SLOWER than 5.05(?) --- wasn't it supposed to be faster or something!To those people who haven't been following the beta's:

DivX 5.1 at performance setting "standard" roughly equals DivX 5.05 at performance setting "slowest" for both speed and quality.

DivX 5.1 performance settings "slow" and "slowest" enable new rate control algorithms that enhance quality, but they also greatly decrease encoding speed. You could compare this to XviD's VHQ and RV9's EHQ features.

So no, the new version is not slower, unless you enable these new features.

[kurdi]

DivX 5.1 at performance setting "standard" roughly equals DivX 5.05 at performance setting "slowest" for both speed and quality.

DivX 5.1 performance settings "slow" and "slowest" enable new rate control algorithms that enhance quality, but they also greatly decrease encoding speed. You could compare this to XviD's VHQ and RV9's EHQ features.


A few questions:
1- Will DivX 5.1 produce a higher quality DivX per CD (700mbs) than 5.05 if allowed to decode at its slowest setting
2- Short of reverting to 5.05, how do I set DivX 5.1 settings so that it is equivalent to the speed and quality of 5.05.

Thanks a bunch

[stax76]

Will DivX 5.1 produce a higher quality DivX per CD (700mbs) than 5.05 if allowed to decode at its slowest setting


since it's more than 10 times slower the question should be how much better is the quality, I didn't test yet but I doubt it's worth the effort

[jeremymacmull]

Even if it is much better the sheer amount of extra encoding time needed on slow and slowest is not worth it in my opinion its better to just keep with two cds on standard than one cd on slowest as it would be better quality on 2 cds anyway and a heck of a lot faster

besides what is the price of paying for one more cd not really worth talking about is it

JEREMY

[Acaila]

1- Will DivX 5.1 produce a higher quality DivX per CD (700mbs) than 5.05 if allowed to decode at its slowest settingAssuming you made a typo and meant to write encode instead of decode, yes DivX 5.1 will produce a higher quality video both objectively (higher PSNR) and subjectively (if you use the new Psy. Enhancements, even though this will lower the PSNR it will look better).
2- Short of reverting to 5.05, how do I set DivX 5.1 settings so that it is equivalent to the speed and quality of 5.05.
As far as I know just setting the performance to "standard" and disable Psychovisual Modelling (because that's slower than it used to be).


As to whether or not the extra quality is worth the slowdown for you is an individual choice, just like with the other codecs.

[stax76]

would be great some new codec comparisons would come up, honestly I don't have the patience to test something that slow myself

[crusty]

I think Doom9 is quite happy to wait until the general mpeg 4 scene warrants a new comparison, but since I can't see in his mind there's no telling what he'll do.

[colordog]

So, I thought I'd throw in my 2 cents worth of observations -

I started an encode last night of two Farscape episodes - they have excellent clean sources, lots of motion, panning, and rocking, and also needed to be IVTC'ed.

Set up DixV5 for "Slower" and PsycoVis to "light". Made each ~45 minute episode 350MB, B-frames but no GMC, 448 x 336 pixels, 128kpbs stereo MP3 (I know, a bit 'old school' on the settings, but I like to keep my disc count down, and watch my stuff doubled in size or full-screen, and find personally that high-quality small blows up better than lesser quality better - just my opinion). Standard IVTC, "light noise" also checked in GKnot 0.28.5.2, and aimed for just two passes.

Anyway, did a 5% comp check. Usually for Farscape episodes in these settings, I'd get around ~80%. With the new settings I got 20-30%!!! I was floored! Indeed, I did another comp check on "Slower" with PhycoVis turned off, and it went up to around 40% on comp check. And it was slow. My megalithic P4 machine would do around 32 fps under these conditions with DivX 5.0.5, now it seemed to crawl around 8 fps. Ugh!

Regardless, I said "what the hell" and ran the encode of the two episodes on "slower" and turned phycovis back to "light" on just two passes. Each episode took a little over 5 hours!!! About 6fps on the first pass, and 9 fps on the second. So, very sucky on the time trade-off here, obviously.

However, after actually running the movie in WMP and expecting to see a ton of low-bitrate compression artifacts, I was treated to the best encodes I'd ever done! Despite the fact that the updated codec said it wanted so much more bitrate, it did beautifully with less. Panning scenes were not overly blurry, foreground faces were just as sharp as background items, etc. Like I said, these are not more compressible files.

So, overall, on my souped-up P4 machine under the conditions above, I take a 4-5x DEcrease in speed, and gain a much much higher quality encode. I never would have imagined settings giving a GKnot compressibility value of ~30% looking excellent. So, I'm excited and disappointed all at once.

If you guys want, maybe I can upload a slice of a clip of one of the episodes for you to prod at.

[QuadraQ]

Originally posted by colordog

If you guys want, maybe I can upload a slice of a clip of one of the episodes for you to prod at.

I know I'd be really interested. I haven't yet seen anything encoded with the new options. I'd like to see how it turns out, and have something to compare it directly to.

[eMMo]

Yeah, a slice would be nice.

[colordog]

Actually, I see that I'm not allowed to post attachments in this forum - anyone know where I might be able to upload a clip then?

[jggimi]

Forum attachments have been damaged for several months. I'll PM you with an e-mail address to use, and then post it on your behalf.

[colordog]

Hey guys - sorry for not uploading the promised clip earlier. Had too much to deal with in life proper.

Anyway, I just finished uploading a clip to our moderator, so hopefully it will show up soon. Be honest with feedback! It's not the best encode certainly, but I think it's great for a high-motion 45 minute IVTC'ed show that fits in 350 MB.

[jggimi]

I have placed Colordog's sample on the web for download. It will be available for the next several days.

To download Colordog's 1.8MB sample, click on (link no longer active).

[len0x]

well, looks nice indeed, but I'd like to compare it one-to-one with 5.05. Colordog, can you do the same encode(meaning resolution and options) with 5.05 and three (or even four) passes and compare?

[colordog]

@len0x

That's a really good idea. I can't do it right away due to other responsibilities, but I'll try to do it real soon.

@all

Now that I've had time to encode one or two dozen items, I believe that I'm starting to see a loose trend. With the settings that I use, I think that 5.1 works much better with clean, clear content - like digitally-shot sources. Conversely, I think it's worse than the previous version on handling 'unclean' content (e.g., film sources with noise, old anime, etc.).

Like I said, it may be very dependent on my settings, but clean stuff looks better than ever, while noise stuff wants to eat more bitrate.

[len0x]

Originally posted by colordog
[BLike I said, it may be very dependent on my settings, but clean stuff looks better than ever, while noise stuff wants to eat more bitrate. [/B]

This can be easily explained by new psycho they have in 5.1

And thanks for the tips - I guess I'm not gonna use 5.1 that soon since most of the stuff I encode is old noisy b&w movies :)

[colordog]

Yikes! Sorry to hear that. Although, I've had pretty good luck using the "movielq" setting with the Convolution3d plugin for old black and white stuff.

I'm glad that it makes sense to you - I'm still trying to learn what really PNSR means for encoding, and the definition of "compensated frame", etc.

[mazzo]

Sorry guys, I know you hate these "what is best" questions. At first I used DivX 5.0x to encode, but switched to XVid and got nice results although the encoding was slower. Now, should I switch back to DiVX 5 again because of the 5.1 release, which the DiVX guys talks so highly about?

[len0x]

Just wanna share some observations about new codec.

First setup: movie source can be considered as a bit noisy (well, I'd say considerably noisy if it wasn't for old b&w movies comparing to which the source seems to be quite clean:) ). So in other words not ideal (like most of non-recent movies actually). 608x336 resolution which gave me ~67% of comp test with bitrate ~950kbps. So while it's still quite challenging for the codec it will be able to produce nice results. (if it was more than 80% then comprasion of different codecs/modes is pretty much useless).

I was comparing three encodes:
- 4 passes with 5.05
- 4 passes with 5.1 (normal speed)
- 4 passes with 5.1 (3 passes on normal, last on slowest)

In all cases no pro options were used (i.e. no b-frames indeed), and no PSY as well. (oh and lanczos resize was used, so pretty sharp picture was expected)

First I have to say that 5.05 performed worse that other two encodes in almost every single aspect. Having said that - I have to admit one thing: if you look at the edges of objects in all three encodes - they are absolutely identical (in both slow and high motion scenes). Which means actually that even when viewing video very close to your eyes most ppl will not be able to tell the difference between those three encodes.

Lets go and do the science now :) What was different and why 5.05 lost? Texture! I have to say staight away that everything that I'm gonna describe below is only visible when doing frame-by-frame comprasion. So 5.05 had fewer detais, more blocks and in general feels like less sharp. So lets move to divx5.1 and see if it's worth using slow/slowest modes of it.

First type of scenes I was looking at was non in-focus scenes (i.e. high motion, special effects etc.) Slowest mode performed well here as I was expecting. Picture looks more smooth and almost don't have any blocks whatsoever. Normal mode gave me slightly visible blocks mostly on such objects like fog, explosion etc. So far so good.

Now I was looking at steady in-focus shots to see how much details codec left in both cases. This came out as somplete surprise to me! Normal mode had significally more details resolved on the picture that slowest mode! (and this actually you will be able to see quite often when watching video as you have time to observe the details of the shot). So in fact slowest mode acted more like PSY (i.e. essentially noise removal filter). May be they have some kind of PSY always enabled in slowest mode...

Anyway imho steady shots are more important (coz they are more visible) and normal mode of divx5 is somewhat closer to what XviD users used to have. And of course when you consider the speed of slowest mode - it don't really see the point of using it at all.

So to sum up:
- Is it worth upgrading to 5.1? Yes, at least for the speed (I forgot to mention that 5.1 normal was faster than 5.05 simply because MV feature is finally working properly).
- Is it worth using slow/slowest mode of 5.1? No. Not only it's slow, but the result are not that great, sometimes even worse than normal mode.

All above is my personal opinion based one one concrete example. Others may have different observations :)

[colordog]

@len0x

Earlier I mentioned that I thought with the new DixV 5.1 and PV that clean (esp. digital) content looked better, while noisy, grainy sources compressed worse. You had mentioned that that made sense based on what you know about PV.

Do you think then that if encoding a noisy source, that it would be better to just disable PV for that encoding?

[len0x]

Originally posted by colordog
Do you think then that if encoding a noisy source, that it would be better to just disable PV for that encoding?

I think yes, and then have some filters in avisynth to do the job. (obviosly PV of divx it's not up to the task of cleaning up source properly...)

[colordog]

So, how did PV change so drastically between 5.0.5 and 5.1? Why is the PV in 5.1 such a potential hog now?

[len0x]

I never really used PV... And I don't know exactly what they did for 5.1, but what's the concept behind PV? It basically herouistic algorithm for removing parts of the picture which is non-visible to human eye. Sort of like noise removal filter, but not quite since you usually have much more parameters to consider. My guess is that guys that designed it were testing it on cleaner sources and got good results, but simply forgot to check noisy ones (or decided to hardcode parameters for the algorithm which is more suitable for clean source).

Anyway it's sort of a blackbox without even any customization, that's why I don't like the idea of using it (when you use noise removal filter you know how it works and can tune it for your needs, but this is not the case with DivX...)