Cvs 2002-02-21

[-h]

I just added some code to disable lumi masking in first pass (there's now no way to enable it - even if you check the box it's forced to be off). Also, the bit bucket system has been changed, so undersizing issues should be gone.

Very small fix overall, I'm working on... stuff.

-h

[Doom9]

let me know when you found my bug.. I'm just doing another pass, this time min quantizers set to 2 and lumi masking off but I still only get quant6 as min.

[Nic]

Get it at:
21-02-02

www.freewebz.com/xvid
or
xvid.stormpages.com

-Nic

[-h]

let me know when you found my bug.. I'm just doing another pass, this time min quantizers set to 2 and lumi masking off but I still only get quant6 as min.

I can't be sure if your bug got fixed or not, the test data I was playing with is back at work (but it helped isolate the issue).

I just uploaded a compile of the latest dll to http://briefcase.yahoo.com/hstink (careful, it's got interlacing code inside as well, though it should be lying dormant). If your undersizing issue is gone... at least I've fixed the right thing :)

Edit: I dare say you'd be better off with Nic's build.

-h

[Koepi]

I guess it's not possible to add my mods to the new cvs - if you changed the core I can't use Isibaars files so the only really "left over" thing would be gone, the frame based quant_type modulation...
But anyways, nice work -h :)

Regards,
Koepi

[-h]

I guess it's not possible to add my mods to the new cvs - if you changed the core I can't use Isibaars files so the only really "left over" thing would be gone, the frame based quant_type modulation...
But anyways, nice work -h :)

We should see a new api in "3 days or so", which will (among other things) consist of setting bits on or off in a "flags" variable, instead of all these separate paramters being sent around in PARAM structures. Once that shows up... I can shove in more of your code! :)

Plus Foxer's code looks fun, and seems a nicer way of handling 2-pass encoding.

-h

[Koepi]

Foxers code is even more nandub-like than mine :)

Now we need a variable (if it isn't there already) with which we can influence the grade of curve-smoothing...

I'll look into it when the new API is out.
Prolly next week - I'm driving home to my parents this weekend, some kind of "family party" (and after that much time for coding again :) )

Regards,
Koepi

[Doom9]

using nic's compile the overflow stays within reasonable amounts for now, plus I saw some quants lower than 6

[duartix]

using nic's compile the overflow stays within reasonable amounts for now

Define reasonable amounts, please.

[Doom9]

success :) desired size 642000KB, achieved size 641994KB. 6KB off, that's what I'm talking about ;) Now I just have to talk vobsub into displaying my translated subs (since the upgrade from v2.08 to 2.09 I haven't been able to get any srt subs via vobsub.. maybe I should try ogg subs again)

[LotionBoy]

you might want to edit those numbers there :-)

LotionBoy

[rui]

Well, i must be getting dumm, or something.:rolleyes:

But, using Nic's latest build, when configuring the first pass, i still can enable Lumi Masking.
Or is it that, even when i enable Lumi in first pass, isn't really enabled in the codec? (in spite that i see it enabled in the tab)

[-h]

But, using Nic's latest build, when configuring the first pass, i still can enable Lumi Masking.
Or is it that, even when i enable Lumi in first pass, isn't really enabled in the codec? (in spite that i see it enabled in the tab)

You can still enable the checkbox, but I guarantee that it isn't being enabled in the codec. I'd disable the checkbox, but enabling/disabling controls between tabs is a major pain, it's really not worth the effort. Better to just confuse people by leaving everything enabled :)

-h

[wing1]

@-h and Nic

I have noticed that CVS 2-21-2002 yield slightly less sharpen video then 2-18-2002 build. I also noticed a slight increase in bitrate (+10K/S) during capture. The observation is comparing the two builds using the same exact settings as it were in 2-18-2002. I never use lumi option and always do 1-pass CBR.

I always do a register edit + add/remove program in control panel to remove the older xvid from the putter. Once that is done I'd do a search in C:\winnt\ for xvid.* and delete it from the system. Next I would reboot and then install the new build.

I also have an issue with the xvid.ax DS filter, WMP's search slider is running and so is the time counter; However, I am watching a black screen and no audio. This is just a simple encoded xvid to AVI format using Vdub. The fourCC are encoded as xvid. If i unregister the DS filter from the registry, I get video + audio. If i registered the DS filter, then I get the problem. I checked file properties to see whether divx DS filter is the problem; However, I did not see divx DS filter for the xvid 4CC encoded vid.

[Nic]

A very quick fix for that -h is:

in
BOOL CALLBACK config_proc(HWND hDlg, UINT uMsg, WPARAM wParam, LPARAM lParam)

Right under:
case WM_COMMAND:

put:
EnableWindow(GetDlgItem(hDlg, IDC_LUMMASK),
psi->config->mode == DLG_MODE_2PASS_1 ? FALSE : TRUE);


That would disable it very easily :)

-Nic

[-h]

I have noticed that CVS 2-21-2002 yield slightly less sharpen video then 2-18-2002 build. I also noticed a slight increase in bitrate (+10K/S) during capture. The observation is comparing the two builds using the same exact settings as it were in 2-18-2002. I never use lumi option and always do 1-pass CBR.

Nothing changed with regards to CBR mode, and certainly nothing changed inside the codec core itself. In fact, only 4 lines of code changed, and they're only to do with lumi masking in 2-pass 1st pass, and 2nd pass quantizer generation.

Are you sure it's not the placebo effect creeping in? :)

I can give you a "10 KBps smaller guaranteed!!" build if you want ;)

I also have an issue with the xvid.ax DS filter, WMP's search slider is running and so is the time counter; However, I am watching a black screen and no audio. This is just a simple encoded xvid to AVI format using Vdub. The fourCC are encoded as xvid. If i unregister the DS filter from the registry, I get video + audio. If i registered the DS filter, then I get the problem. I checked file properties to see whether divx DS filter is the problem; However, I did not see divx DS filter for the xvid 4CC encoded vid.

No idea.

:: passes on to Nic :: :)

-h

[-h]

A very quick fix for that -h is:

Ah.. if only it were that simple ;)

I have no idea why, but enabling/disabling controls between property sheets is a royal pain, you pretty much have to create a thread to enable/disable things a few milliseconds after the page has loaded to get it working properly. Your way would work, but only after going to the Advanced tab twice - the first time, the change wouldn't take effect.

Not without some "coaxing" anyway :) There are other ways around it (i.e. calling config_download() on combo box update, plus some other mojo) but it's all a lot of trouble.

-h

[Nic]

But it is that simple :)

Try it...it works fine :)

-Nic

edit:
@Wing:
Could you try jumping around the file (to see if it ever starts up). Seeking without many keyframes can be troublesome (especially in DivX4). Are you using my DShow filter (with brightness) or the normal one.

[-h]

But it is that simple :)

I can't believe it!

Nay..

I won't believe it! :)

-h

[Nic]

Try it....damn you, try it. :)

-Nic

[-h]

Try it....damn you, try it. :)

Agh.... it's so so much easier when you keep up the ruse of "it's too hard to implement!! let's do something else!!" ;)

The next CVS update should be more fun.. Foxer's curve mods are very nice, after I finally got to test them.

-h

[Nic]

Agh.... it's so so much easier when you keep up the ruse of "it's too hard to implement!! let's do something else!!"

Wont fool me though :)

Im looking forward to the next update, and to getting post-processing working.....

....Do you know anything about the quantizer store? It was used in the old decore alot, but I dont know exactly what it is referring too.

-Nic

[-h]

....Do you know anything about the quantizer store? It was used in the old decore alot, but I dont know exactly what it is referring too.

Hrm I haven't looked at the old decore at all..

I assume it's an array of quantizers used for each macroblock? If so, core could supply that quite easily..

-h

[Nic]

so the array of MACROBLOCKS have a dquant value.....Do you reckon that could be it?

-Nic

[-h]

so the array of MACROBLOCKS have a dquant value.....Do you reckon that could be it?

The dquant value is the difference in quantizer between the last and current macroblock. I assume decore is relying on absolute quantizers, in which case you'd have to start with the initial frame quantizer, and add all the dquant values in sequence to get the actual macroblock quantizers that it's after.

Of course, certain encoders (DivX 4.x, for one ;)) don't ever use dquant values, since they don't do quantization changes on a per-macroblock basis. I'd have to look at the source more to be certain, but I'm still pretty wasted after a night out.

So tomorrow would probably be better :)

-h

[Nic]

Oh, I missed the "quant" in MACROBLOCK, that looks a bit bit more like it.

-Nic

(Im off to the Dog Racing tonight, so ill probably feel worse for wear soon (no Foster's there though, just lots of Carling))

ps
& I can get to the Macroblocks through dec->mbs :) So as long as that is the quantizer store referred to...im happy :)

[wing1]

@-h

hmm, That is very strange. I've installed & uninstalled the two builds over and over to compare the two. It could very well be me and my imagination again :)

10K/S is not that big of a deal for file size anyway so I not worry much about it. It just my observations between the two builds. I'll go back and do some more observations tonite after work, and this time i'll do the observation with beer instead of whirlwind tea and see if that would make a different to the sharpness in the vid :)

@nic

I did move the slider back and forth to see if I can get some sort of a response from WMP; Nevertheless, all I get is black screen and no audio. The strange thing is that the very same file will come to life as soon as I unregistered the xvid.ax. I am testing with both the included xvid.ax from the 2-21-2002 CVS and the separate xvid.ax from your link that has the brightness support.

I always use divx4.12 filter as suggested by -h and Doom9 on xvid playback; However, when I saw the new DS filter for xvid naturally I wanted to try it out and see what's new. So I encode a 10min clip with 2-21-2002 CVS build using xvid 4CC and test it. I've yet tested with older CVS build, and I will do that tonite. I've used xvid DS filter before with 2-16-2002 build and it work fine, only this one that I am having issue with.