ffdshow development

[Alxemi]

Ligh: You are right about the spam, but i´m not worried about it, because of the REMOVETHIS a robot cannot use the address, it must be somebody that write it down.. The funny thing is that i think i sent the private message but it was late at night here... and you kinow, sometimes you don´t remember the things you have done just two minutes before xD, anyway thanks for the advise; i will remove the address just in case :D
With your mirror is not necessary another one but now that the work is done...
http://www.alxemi.cjb.net//d9samples
(the double slash is ok)

Regards

[LigH]

I wonder if anyone can confirm the "orange-blue" colorisation bug; it may only occur on some special material, I could not yet create a DivX file that showed the same effect, but could not yet try all encoder options.

For some strange reason I think that there is a tool which can tell details about encoded DivX-5 AVIs (which codec options were set upon encoding); am I right here?!

[Soulhunter]

Yet another feature request...

Could something like this (http://forum.doom9.org/showthread.php?s=&threadid=63361) be added to ffdshow ???

I mean the Unsmooth(AddGrain x,x,x,) method ! ;)

Maybe, regular Unsmooth() with different options too... !?!

Would love it, to see this thing in ffdshows filter-tab ! :D

Bye

[LigH]

Still no interest in resolving the orange-blue bug? Is this the wrong place to post bugs, shall I contact someone else somewhere else?!

Okay, for your pleasure, one more image - this time a bit larger, but still only a cropped part (due to the explicit content of the rest):

http://www.ligh.de/private/nikita.jpg

And finally, I cut out some non-hardcore parts of the mentioned movies for download; but to protect my bandwidth, please contact me if you want to download them (1.5, 2, 3.5 MB). And remember: The effect is only visible using ffdshow Beta 2003-09-27.

BTW: I was absolutely unable to recreate this effect - I tried all combinations of options I could think of, but DivX 5.0.5 for the PC did not create one file causing the same effect. I'll have to contact the creator of these movies and beg for supporting us by telling us the details.

[esby]

A dummy question...
Can ffdshow intecepts keys in its state of directshow filter?
-->
Toggling OSD display...no matter the player you are using
etc.
And allowing custom key code supposing no application is already intercepting them...

On another note, the osd could be customizable
like a sprintf sequence ( ...,"now displaying frame n°%i" ,nbframe)...

esby

[Soulhunter]

@LigH

Sorry, but have never seen something like this before... :confused:

Are you sure that this is not a source related thing ??? :(

Bye

[LigH]

Well, what do you mean by "source related"?

- Only those movies from that creator show such an effect, I could not yet find any other movie with the same problem (but I'll search for more).

- Only ffdshow 2003-09-29 Beta shows such an effect, older ffdshow filters or the original DivX 5 filter does not.

You see, it's only this combination; but all the movies from this creator (which I collected over months now) show the same bug, more or less obviously depending on the content, but safely reproducable.

[athos]

Did you try different DCT?

[LigH]

Didn't I write that?

It does not depend on the selection of the iDCT.

I just wrote an email to the creators. And I have a few URLs for downloading video snippets for you. I'll send you a PM so you can download them and check. Also I saved the ffdshow configuration to an ".ffpreset" file.
__

P.S.: About my system:

- AMD Duron 800
- ABIT KT7-RAID
- ELSA Gladiac GeForce2 GTS, 32 MB
- ELSA DetonatorX (6.13.10.4109)
- DirectX 9.0b (4.09.0000.0902)
- Desktop: 32bit TrueColor

[Tommy Carrot]

It can be GMC related bug too. It's rarely used, maybe the implementation in divx5 is not totally compatible with the standard.

[LigH]

But I created movies with all the 8 combinations of Pro features, and neither of them created a movie which showed the same effect.

Well - it could indeed be possible that this movie was not created using the DivX codec, but instead got the DX50 FourCC. Maybe it's XviD with 3 warp points, but ffdshow believes it is DivX 5 and only expects 1 warp point?!
__

Well, at least this is probably not true:

- The original AVI has the FourCCs "DX50/DIVX". (DShow/VfW)
- If you select "DX50" inside the XviD codec, it would have "DX50/xvid".

Furthermore, I just tried to recompress with XviD (latest Koepi build) with GMC and "DX50/xvid" FourCCs: ffdshow played fine.

I even changed the FourCCs to "DX50/DIVX": Same result, no problem.

[Danzel]

If you recompressed it with XviD then that would of 'fixed' the problem, as your original stream would of been decoded with vfw and then re-encoded with xvid. So it wouldnt be the same stream anymore.

Have you tryed to change the FourCC of the Original Files to have XVID FourCC ? (without recompressing, just changing the FourCC).
And Then playing them then with FFDShow?

It could be its expecting different GMC, although I thought that information would be stored in the stream anyway.

Perhaps they are encoded with a buggy DivX 5?

Danzel.

[CruNcher]

you could get the mpeg4 bitstream id to see wich encoder was used so you would have a clearer picture of the situation instead of trying to guess what was used by looking @ the riff header (fourcc) use http://www.geocities.com/filesDRF/DivXDRFAnalyzerV0951.zip for example
for me it looks like a problem @ the encoder level even if not much can be seen of the problem :rolleyes:

[LigH]

@ CruNcher:

Thank you very much - that's a kind of tool I missed a lot, I hope to see something clearer now!

(Unfortunately, the GeoCities link is off, but I found a few via Google: Lots of FTP links...)

{ EDIT: The GeoCities address is working, but you have to visit the homepage first, else download is blocked (anti-leeching). }

DivX DRF Analyzer v0.9.5 Report!
File Name: N:\Movies\ALS\nikita001.avi
FourCC: DX50
Codec: DivX501b484p
Resolution: [ Width: 320 Height: 240 ]
Frame Rate: 29.970 frames per second
The Video has 2997 frames [ 00:01:40 ]

Average Frame quality is HIGH [Average DRF/quantizer is 2.99]
Standard Deviation: Quality is MEDIUM [Std. Deviation is 1.01]
Image Resolution is LOW

The file has Packeted Frames!

There are SOME frame drops ( NO drops is better )

Recomended Resolution: [304x224] (Target DRF/quantizer=2.8)
The filesize should be larger!

Performance Caracteristics:
Macroblocks per frame: 300
The Width is multiple of 32

Kilobits per Second: 642.38
Kilobits per Frame: 21.43
Kilobits per Macroblock: 0.071
Bits per Pixel: 0.29

Frame Type Statistics :
I Frames: 0.80%
P Frames: 20.05%
B Frames: 49.15%
S Frames: 29.96%
N Frames: 0.00%
(More Advanced Codecs use B and S frames)
Frame Quality Statistics :

DRF=1&2: 1496 50.3%
DRF=3: 1 0.0%
DRF=4: 1472 49.5%
DRF=5: 0 0.0%
DRF=6: 1 0.0%
DRF=7: 0 0.0%
DRF=8: 2 0.1%
DRF=9: 0 0.0%
DRF>9: 0 0.0%
KeyF/DeltaF: 0.81%
KeyDRF<4: 23
KeyDRF=4: 1
KeyDRF>4: 0

AverageKeyDRF: 2.08
MAXDRF: 8
AverageDRF: 2.99
Deviation: 1.01

http://www.geocities.com/analyzerDRF/

Obviously, this must be a bug from a rather old DivX codec version. Are there any "rather complete" changelogs of DivX codecs available?

{ EDIT: http://www.divx.com/divx/divx_win_versions.php }

[BoNz1]

Anyone played around with the sorenson3 decoder? It works really nicely, I got the new high def Matrix Revolutions trailer to play in graphedit w/ 3ivx parser, CoreAAC, and ffdshow. Sweeet! The audio goes out of sync but I think thats the 3ivx parser. Anyway, the video looks awesome.

[LigH]

Obviously, several rounding bugs were solved towards DivX version 5.0.3 - maybe it was one of those. I could try to install DivX 5.0.1 over the weekend and check if it produces such "provoking" movies.

Until now, no answer from the creators, but as we could find out the version number using that analyser, we probably won't need their support anymore...

[CruNcher]

@LigH

hehe who were the creators i would guess KRC amateurs ;)
ups a nevermind even 320x240 is under KRCs niveou and i was kind not
to talk about those droped frames :P

[LigH]

:confused: What is "KRC"? Probably no popular abbrevation in Germany?

These are no cinematic movies - everything is original content, taken by hand cameras; so I can guess the reason of drop-frames. BTW: It is adult content...

[CruNcher]

BTW: It is adult content...


i never would come on that idea myself from you previous behavior:rolleyes: *looks arround* so this is privately recorded no professional stuff ? eh i think we better cut off that discussion now or doom9 will hang us :D *cough* nevertheless you could pm me the address where i can get this sample to take a look @ it eh the problem ofcourse even if it's solved ;)

[Lobuz]

@Athos
Any chances for new compile of ffdshow?

Regards
Lobuz

[LigH]

@ Lobuz:

Please remind that before a new ffdshow can appear, Milan Cutka or the ffmpeg developers must have been active; and after discovering several quite important problems, it may take a little longer to investigate and solve them.
__

@ Athos & others involved:

Did someone contact the useful addresses about our latest results, or shall I do that (then I'd like to be told where)?

[athos]

New builds are up:

ffdshow-20031028 (http://athos.leffe.dnsalias.com/ffdshow-20031028.exe) changelog (http://athos.leffe.dnsalias.com/ffdshow-changelog.txt)

ffvfw-20031028 (http://athos.leffe.dnsalias.com/ffvfw-20031028.exe) changelog (http://athos.leffe.dnsalias.com/ffvfw-changelog.txt)

NOTES:
I had to reinstall everything due to a crash, and believe me there are lots of SDK's and stuff to install and configure for this! Problem is: I could not compile libavcodec or mpeg2enc using gcc/nasm, and so I had to compile them using Intel C++ Compiler 7.1 with every imaginable optimization flag ;) This means they lack assembler optimizations, put perhaps the ICL can somewhat make up for this. On my system (P4-3.0Ghz) it doesnt seem to make too much difference on a quick check, but I suspect it will on slower systems.

Anyways: enjoy!

[Gaia]

Thanks :) Now you can enable overlay mixer.

Edit: You can't enable it. It still doesn't load if you enable it...

[winman]

@athos

Thank you very much for the new build.

I can confirm a substantial performance hit with the new build of ffvfw (atleast on my overclocked 2.2Ghz Athlon XP system).

Here is the result of a quick test of a 640x480 video.

MPEG4 One-pass CBR 10000Kbps with all extra option like ME off

62-99 fps 20030927
23-44 fps 20031028

With Full ME,Trellis and extras...

10-15 fps 20030927
1-2 fps 20031028


My favorite feature of the recent ffvfw, the speedy ffmpeg MPEG2 encoder, also take a similar perfomance hit. I hope you can get gcc/nasm working again because I really love the 2X-3X real-time MPEG2 encoding :)

[athos]

i will try to get it to compile using gcc/nasm, it is most likely something i misconfigured or so. my test was decoding using ffdshow, which doesnt take more than a few % cpu anyway.

[Mr.Shine]

Originally posted by Gaia
Thanks :) Now you can enable overlay mixer.

Edit: You can't enable it. It still doesn't load if you enable it...

Yes, there is a problem with overlay in ffdshow, since the first build where the "Use overlay mixer" checkbox started to have 3 states: checked, unchecked and grayed. In checked state, WMP will fail to render the file, the other 2 settings (unchecked and grayed) will disable overlay in WMP (I tried 6.4 only).

There are "cheats" to get it to use overlay, though. Uncheck "Use overlay mixer" in ffdshow and do one of the following: add a filter that installs itself after ffdshow in the directshow chain and allows for enabling overlay (e.g. DivXG400 in Extended Overlay mode - disadvantage is that DivXG400 only supports YUY2, so there will be an unnecessary color space conversion). This should work with any player, including WMP.
use BSPlayer and enable overlay there.
use MPC or ZoomPlayer and specifically tell it to use the Overlay Mixer instead of "System Default".I don't know what happens if you try to use VMR or VMR9, as my display drivers don't support VMR.

[Blight]

Athos:
It this another interm release or some sort of semi-authorized release?

I for one would like to see a feature freeze and see some bug squashing take place and a non-alpha/beta release out the door.

Right now FFDShow is in a state where some builds have some features working, while other builds have other features working and I can't even guess as to which version to recommend to users.

[LigH]

The Orange-Blue bug was not yet fixed; I think I should try to send a mail directly to Milan.

[Gaia]

Originally posted by LigH
The Orange-Blue bug was not yet fixed; I think I should try to send a mail directly to Milan.

Are you sure it's ffdshow bug? Not some problem in your amateur xxx movies? Because you're only person having it :)

[Danzel]

If you found which version of Divx was used to encode those clips then it would make testing a bit easier (and allow for g rated tests ;))

From the info that program gives you should be able to work out which version it was encoded with, perhaps we could even ask the DivX peeps if they know of any errors that they may of found that could of caused this problem (If its encoder related).

Danzel.

[athos]

Originally posted by Blight
Athos:
It this another interm release or some sort of semi-authorized release?

I for one would like to see a feature freeze and see some bug squashing take place and a non-alpha/beta release out the door.

Right now FFDShow is in a state where some builds have some features working, while other builds have other features working and I can't even guess as to which version to recommend to users.

Well I guess you would call it interm release. I asked milan, and he does not want me to put up these build on sf, but he does not have anything against me releasing them "privately". So they are in no way official, even as "alpha". Milan is in the middle of integrating ffvfw with ffdshow, and he also wants to get the colorspace conversion working the way. When this is done, I think he aims to put out a new release on SF, ie "official".

[oddball]

I am going to wait for official since Athos (No offence) your builds have jerky video.

[oddball]

One other thing. Someone mentioned something about G400 and colorspace. Would using DiVXG400 acount for the 'greening' problem when playing back XviD with libav? If I change it from libav to XviD the greenout problem goes away.

[LigH]

:confused:

@ Gaia:

The only one? athos downloaded my samples and could confirm it. If you want to check it too, write me a PM to confirm that you really want to watch such content... ;)

As I already told a few times: This bug only appears in this combination:

- ffdshow 20030927 or 20031028 (any other filter able to decode DX50 I have is not affected)
- movies from this source (maybe also others, but I could not yet find any more)
__

@ Danzel:

Would you be so kind as to look at the top of page 19 of this thread?

DivX DRF Analyzer v0.9.5 Report!
File Name: N:\Movies\ALS\nikita001.avi
FourCC: DX50
Codec: DivX501b484p
__

@ athos:

Discovering such a remarkable change in the decoding behaviour (even if it only regards very few and special files, but unfortunately so obviously) is worth to provide such "interim releases", in my opinion...

[Danzel]

@ligh haha, yeah i saw that.
but which version of DivX actually produced it, is that 5.01?

anyone have 5.01 installed / installer and would be able to see if it creates a stream with the same codec string?

Danzel.

[wannabe]

I downloaded the latest this "ffdshow-20031028.exe". And it still exhibits the -same- XviD iDCT problem mentioned before...

The option in Misc: "autoselect" doesnt work either.

[LigH]

Unfortunately, until today no reply from the creators. So I can only guess: Most probably DivX 5.01; and I'll have to install it to check if I can create more movies with that effect (currently, I have 5.0.5 installed, and there are no orange-blue problems; between 5.01 and 5.0.3, several rounding-bugs in DivX encoder were fixed).

[LigH]

Updates:
yablo found another movie, created with DivX 5.0.2, which shows the same effect. Another hint that just these fixed rounding bugs in the DivX encoder could be the reason...
Milan just answered, and he already has a strong suspect. He will soon be able to download my files and check if he thought right. Then the CVS shall become updated.

[dillee1]

Originally posted by Blight
Athos:
It this another interm release or some sort of semi-authorized release?

I for one would like to see a feature freeze and see some bug squashing take place and a non-alpha/beta release out the door.

Right now FFDShow is in a state where some builds have some features working, while other builds have other features working and I can't even guess as to which version to recommend to users.

authorize or not doesnt applies to GPLed stuff,
as long as

- GPL is complied
- No law/patent is broken in his country due to distro of the binary.

The whole idea of GPL is everyone can get a hand of the source and tailor it to whatever he likes..... Apart from may be it's a "manner" to notice the original author about this, why would we care about authorisation?

[eLupus]

What he meant was that these compiles are not compiled by the developer and have not been put up as a binary release on sourceforge. This is cause the code is going through major changes and Milan doesn't think t is ready for public release. This doesn't hinder anybody from compiling the source from cvs themself, but they should then be aware that it might not even be possible to use.

Regards
Joakim Plate

[vlada]

Hi, does anybody else have problem with the last 2 unofficial releases of FFDShow? If I install it and playback a movie, the player want use the FFDShow decoder, but any other asociated. If I make the player to use the FFDShow decoder, it crashes. The version from 2003/08/16 works O.K. What could be the problem? Thank You, Vlada
P.S. I have WinXP SP1/ATI 9200/Catalyst 3.7/DirectX 9.0b, I belive a pretty standard computer.

[gabest]

I just noticed yesterday evening that the latest from athos's site was broken. It set some crazy output resolution and the renderers refused to accept it (fortunatelly).

[Kyo]

Vlada

Just disable the "Use overlay mixer"

[Owen]

"Use overlay mixer" is off by default.
Why would anyone turn it on ???

ffdshow 20030927 works great. Ffvfw is comming along very nicely as well. Many thanks to the great Milan.
Thanks also to athos for compiling and hosting the builds.

I look forward to the next official release.


Owen

[vlada]

Overlay mixer - this will be probably the problem. I checked it, but to be honest I'm not sure what it exactly does. I belived, it uses overlay mixer to view the movie. I'll try to uncheck it and let You know if it was the problem. Tahnx, Vlada

[Wilbert]

People activate it when playing matroska files with a certain aspect ratio (other than 1:1).

[QQ]

yes, I have same problem.. newest builds seem to crash when trying to play anything...... i have no idea whats the problem..

[QQ]

ok, I found the problem.. if i disabled Nic's post-processing method, it doesn't crash anymore..

[QQ]

overlay mixer doesnt work though..

[Soulhunter]

@QQ

Ehm... :rolleyes:

Do you know about the function of this funny little "edit" button down there... ???

Bye