ffdshow development #2

[bond]

ok, now that, after more than one year of development and testing, a new build of one of the best mpeg-4 (and more) directshow decoders existing, ffdshow, has been officially released, i take this opportunity to close the good old 65 sites strong ffdshow development (http://forum.doom9.org/showthread.php?s=&threadid=48511) thread, as its simply not possible anymore to find any infos in it

everyone is invited to continue the discussion about ffdshow development in this thread or start new threads for discssing specific issues in seperate threads :)

grap ffdshow here (http://sourceforge.net/projects/ffdshow/)

[Soulhunter]

@ Coroner

Thanks for the info !!!

Only asked coz there is this winamp plugin...
Version 1.2

SBLive! Resample BugFix output plug-in for Winamp2

Freeware component!

USE THIS SOFTWARE AT YOUR OWN RISK !!!

I will not be responsible for anything regarding the use of this software!

If you are not looking for highest audio quality or have no idea about what
distortion is, than this component is of no use for you ... it will only make
your CPU work more on music

but for those with SBLive! or Audigy series of soundcards and there are a lot more
with this issue! this will give you the perfect sound you wanted to hear when
you've bought that soundcard !!!

[The SBLive! issue]

SBLive! has a low quality resample to 48KHz algoritm (or multirate filter
kernel too small...) maybe because to keep low latency to audio stream ...

This is the SBLive! (and not only) problem:

absolutely any audio stream is converted to 48KHz.
the worst performance is 44.1KHz audio streams (cd audio).

Audigy has the same issue except it resamples to 96000Hz but still
kernel size too small so 1% IMD at 16KHz for Audigy2 (www.tomshardware.com)

[The solution:]

my high quality resample to 48KHz (or any >= 48000, that means 96000Hz too)
component for Winamp3.

Testing results:

using SBLive! to resample to 48000:
19KHz IMD=20%
18KHz IMD=10%
16KHz IMD=5%
14KHz IMD=1%

You don't believe me ...???? I've included in component pack generated tones
used in my test versions of 44100Hz and 48000Hz ... Should sound the same !!!

using my component for KernelLength=16000 samples:
19KHz IMD=0.01%
18KHz IMD=0.001% ; can't measure less than this
16KHz IMD=0%

CPU Load:
<14% on Duron 825MHz for KernelSize 16000
<5% on Duron 825MHz for KernelSize 6000
<1% on AthlonXP2000+ for KernelSize 20000

[known issues]
there are no known issues

<TargetSampleRate> specifies resample target frequency of filter (should be
48000 for SBLive! cards, 96000 for Audigy, and this must be the working
Samplerate of the DSP on your soundcard!!!

<KernelSize> is the multirate kernel filter length in samples of the lowpass
filter. Here are some hints for 48000&96000:
16000 samples for quality described above
6000 samples are enough for 0.1% IMD @ 19KHz and no audible IMD
(IMD is above 20khz and should be rejected by the hardware
lowpass filter of your soundcard)
altering this results:
higher KernelSize => higher quality more CPU load.
lower KernelSize => lower quality less CPU load.
don't alter this if you do not know what means this or if you do not check
the results on a spectrum analyzer!
I've used as spectrum analyzer Goldwave 4.23 and CoolEdit2000 to measure
IMD. To record result just loopback your soundcard output to Line-In and
view the result over SpectrumAnalyzer in realtime.

[some hints]
If you really want your SBLive/Audigy/Audigy2 to sound perfect I recommand you
to download and install KX Drivers wich you can find at: www.kxproject.com.
These drivers are opensource and also freeware.
Best sound : KX Drivers + SBLive!Bugfix plugin. Try it!

[Contact:]
email: adi111p@yahoo.com


P.S. Haven't you noticed that DVD movies (sampled to 48000Hz) sounded perfect
on your sound system and Audio CD's didn't????? This is why!

another advice: Use KX drivers for SBLive and Audigy and Audigy2 soundcards. You will get
professional EQ-s, filters, 5.1,6.1,7.1 Decoders, but you will need some time to understand
how to configure it

download it from www.kxproject.com

So, still need to know this about ffdshows upsampling...

- Is Kaiser "the best" upsampling algorithm of ffdshow ???

- Is it better than the Audigy's "bad" resampling ???

- When yes, how much better is it ???


Tia n' Bye

[unmei]

great it even comes with a built in microwave :eek:

I don't know it this is ultimately new, but it wasn't in the 2004-06-29 i used so far: keyboard shortcuts to toggle PP, noise etc ..good idea, very useful!

[Coroner]

@Soulhunter

Haven't seen that one before. Same trick as resample in foobar by the looks of it. Would avoid the resample at 48Khz, although it's still not going to sound as good as a non-resample card, depending on the resampler it would probably sound better than the creative one. Suggest you might get better info by posting a question here Head-Fi Computers as a source forum (http://www4.head-fi.org/forums/forumdisplay.php?f=59)

Hope that helps

[saratoga]

Any chance of an Athlon or P4 optimized build? Or is that not worth the effort?

[opsis81]

Please fix the broken MPEG-1 and MPEG-2 decoding...
Please make me get rid of any other MPEG-2 decoder I have registered in my PC...
Thank you in advance!:)

[SeeMoreDigital]

Originally posted by opsis81
Please fix the broken MPEG-1 and MPEG-2 decoding...
Please make me get rid of any other MPEG-2 decoder I have registered in my PC...
Thank you in advance!:) I second this request!

It would be great if it could support Mpeg2 HiDef pixel frame sizes right up to 1920x1080/88...


Cheers

[SeeMoreDigital]

Originally posted by bond
... Grap ffdshow here (http://sourceforge.net/projects/ffdshow/) It's probably worth pointing out that the FFdshow builds on the SourceForge web site are by Milan Cutka.


Cheers

[athos]

Originally posted by SeeMoreDigital
It's probably worth pointing out that the FFdshow builds on the SourceForge web site are by Milan Cutka.
Cheers
The ffdshow-alpha builds are actually by me.

Did you notice the new ffdshow homepage that Milan put up?
http://ffdshow.sourceforge.net/tikiwiki/

[bond]

Originally posted by athos
Did you notice the new ffdshow homepage that Milan put up?
http://ffdshow.sourceforge.net/tikiwiki/hm, cant reach it atm :(

[hellfred]

It sure took its time, but finally a side came up in my browser.
So do not give up.
Here is a deep link (http://ffdshow.sourceforge.net/tikiwiki/tiki-view_articles.php)
Hellfred

[SeeMoreDigital]

Originally posted by athos
The ffdshow-alpha builds are actually by me.

Did you notice the new ffdshow homepage that Milan put up?
http://ffdshow.sourceforge.net/tikiwiki/

I apologise Athos. But why is Milan's name slapped all over the change logs?

http://sourceforge.net/project/shownotes.php?group_id=53761&release_id=255667

If this has confused me, I can't imagine how the newbies are going to get on :confused:


Cheers

[athos]

Originally posted by SeeMoreDigital
I apologise Athos. But why is Milan's name slapped all over the change logs?

http://sourceforge.net/project/shownotes.php?group_id=53761&release_id=255667

If this has confused me, I can't imagine how the newbies are going to get on :confused:


All CVS commits are by Milan. I just compile the binaries.

[DAvenger]

Many thanks to Milan for 'fixing' the wildcard issue. Now RadLight users can use FFDShow once again :D

(Milan, to bolo teda rychle hehe :p )

[faxmactor]

Sorry for taking over this matter from the previous thread, but I haven't replied this reply ;), and besides the problem still exists.

Originally posted by faxmactor
I have video that is likely to be made with an buggy, early Xvid 1.0 (or earlier) build. It has lots of b-frames but no keyframes in the beginning.
Xvid is able to display it correctly as well as mplayer (which uses libavcodec, too), so I came to a conlusion that this issue could be fixed in ffdshow as well. Milan, please have a look at.
A screenshot: HERE (http://ozdtersegi.axelero.net/ffdshow_issue.jpg), a short clip (2,5M): HERE (http://faxmactor.port5.com/Excel%20Saga%20-%2001%20-%20The%20Plan%20To%20Murder%20Koshi%20Rikudo%20(AHQ).ogm)
Originally posted by Bogalvator
Did you try setting the IDCT setting to XviD? (under the "Miscellaneous" section)
Yes, I did. No use...

[Andy2222]

new SSE2 release up

NOTE: This is just a preview version, means its not as "rdy" as the older releases since im reworking kinda lots of stuff in the resizer and had no time to finish all.

I released this versions mainly cause milan added all this new audio stuff and this seemed usefull for the SSE2 version too.

PS: read the sse2 changelog first before u update since some small stuff is missing in this preview version!

u can download it from the sig location

[Soulhunter]

Yay, milan added the SRC resampling Ive requested... :cool:

Cant await the non SSE2 version !!!


Tia n' Bye

[Audionut]

Originally posted by Andy2222
new SSE2 release up


Thanks Andy2222 for your sse2 releases.
And to Athos for giving you somewhere to stash your realeases.:D

[Alvy]

Ffdshow really got awesome in its possiblilities :) !
Thank you for your work.

It would be great if the hotkeys settings could get more powerful. So the user could comfortably take control over ffdshow.

For my personal usage it would be fantastic it one could control:

misc:
video delay +/- (10 ms) per keypress

resize:
black borders
+/- per keypress or at least toggle on/off

much less important
zoom +/-
shapen strength +/-

I would be very glad if someone could improve the functionality of the hotkeys.

Thanks

Alvy

[Soulhunter]

Please could someone with a Audigy2 ZS help me to confirm something ???

Do the following steps with the 20040725 build...

- Set your Audigy2 to 5.1 setup

- Disable all extras (EAX, CMSS, EQ etc.)

- Connect a headphone to the L/R output of the card (or a 2ch amp)

- Activate ffdshows raw audio processing (no decoding/filters/mixer)

- Playback some stereo content (a simple mp3 file should do it...)


So, playback should be balanced (same loud @ L/R channel) !!!

When yes...

- Enable ffdshows audio resampling (96000Hz / Kaiser)

Do you hear a difference in the channel balance now (is one channel louder) ???


EDIT:

Addition to the last post...

Plz, try the above stuff with the default integer output (all checked -> 16/24/32) !!!

If the R channel gets louder (ca. 25%) after activating the resampling...

Disable 16bit integer and use 32bit floating point to see if it helps !!!


Tia n' Bye

[Soulhunter]

Ok, forget the stuff above !!!

Problem is solved with the last build... :)


Bye

[omion]

I've been having a problem with the latest (20040808) ffdshow build. It gives wierd 8x16 blocks in most b-frames of most movies. All of my movies are XviD or DivX (a whole bunch of versions... I don't know which ones are which) The blocks only appear when decoding b-frames with the latest build, and decoder set to libavcodec.

I have tested it in MPC and The Core Media Player, with the same results.

I have a P4 1.5GHz with 512MB RAM.

You can see an example here. (http://people.ucsc.edu/~rswilson/other/sakura-vs-ffdshow.jpg) Note that all the things that look like JPEG artifacts are actually JPEG artifacts. :p

Thanks!

[Ana]

I have the same problem with ffdshow (20040808) that omion has, when libavcodec is used to decode xvid. I think this has something to do with custom matricies.

Xvid (1.0.1) setting used to encode:

matrix: Andreas 78er
Qpel
GMC
max b-vobs: 2
VHQ 4
chroma on

[Gooop]

I don't think it has something to do with custom matricies, as I have the same problem on many videos which were encoded using h263.
2 bframes as well, QPEL but no gmc. I don't know if it can help but it seems that the problem is not exactly the same if the video is muxed in a mkv : in avi I can see the same artefacts as omion, but once muxed in mkv the video just crashes my player, mpc 6482 with its builtin matroska splitter. Another weird thing is that the player doesn't crash if I play the same mkv file with the same mpc but another matroskasplitter I registered (matroskasplitter_20040111-2). In this case the video plays with the same bugs as described before.
And I think (not sure, long time ago...) the problem occurs with videos that I encoded with older xvid builds as well (koepi-19062003-1).

[arno]

Is there already somebody out there that has some info on what "Mplayer Accurate Deblocking" exactly does (or has a link with an explanation)? If I interprete it right it sounds like a nice feature but a solid explanation would be nice :-D

[masken]

(repost from closed thread)

Is there any chance of improving the installer so that it supports silent installs? (/Silent switch for example).

...and perhaps also return an errorcode upon failure.

Perhaps also some additinal switches, like:
/DisplayProgress
/PostProcessing:[0-6]
/AutomaticQualityControl
...etc.

Or alternatively, the HKLM keys explained so one could just import values silently after an install to set the preferences.

This would really help me (and many other I guess) when installing new PC's....

[swalker]

Originally posted by masken
(repost from closed thread)

Is there any chance of improving the installer so that it supports silent installs? (/Silent switch for example).


It already does. The ffdshow installer is your standard NSIS based installer which supports /S for silent installations.

[winman]

@omion & ana

I also has the same weird blocks with some XviD and DviX movies. However, after changing the IDCT from the default "simple" to "auto", the weird blocks went away.

Previous version ffdshow-20040725 has IDCT default at "auto".

[omion]

Originally posted by Gooop
I don't think it has something to do with custom matricies, as I have the same problem on many videos which were encoded using h263.
2 bframes as well, QPEL but no gmc. I don't know if it can help but it seems that the problem is not exactly the same if the video is muxed in a mkv : in avi I can see the same artefacts as omion, but once muxed in mkv the video just crashes my player, mpc 6482 with its builtin matroska splitter. Another weird thing is that the player doesn't crash if I play the same mkv file with the same mpc but another matroskasplitter I registered (matroskasplitter_20040111-2). In this case the video plays with the same bugs as described before.
And I think (not sure, long time ago...) the problem occurs with videos that I encoded with older xvid builds as well (koepi-19062003-1).
I have strange things happen with Matroska, too. The screen shot I posted was from a Matroska, decoded with the external splitter (1.0.2.3 + sub patch) I demuxed the video, and tried to play the resulting AVI, but it crashed. Using the internal Matroska splitter, it also crashed.

Again, everything is magically fixed when using either an older version of ffdshow, or not using libavcodec.

I also just noticed that the blocks ONLY occur on the right half of a 16x16 macroblock, NEVER on the left.

@winman:
Doesn't help for me. :( I have mine set at XviD (libmpeg2) all the time.

[Blkbird]

OK, I asked this in the "old" thread, but it wasn't answered, so here I try it again.

Why are only TrueType fonts listed for subtitle display? The Type-1 and OpenType fonts are not showing up in the list.

[ViCroié]

I've been having a problem with the latest (20040808) ffdshow build. It gives wierd 8x16 blocks in most b-frames of most movies. All of my movies are XviD or DivX (a whole bunch of versions... I don't know which ones are which) The blocks only appear when decoding b-frames with the latest build, and decoder set to libavcodec.

I have the same problem, i tried both H.263 and Custom (HVS) matrice... but the only way for me to not have these blocks was to disable QPEL... when i use XviD's own decoder... it plays fine... for some encodes i used XviD with VHQ for B-Frames, but i tried XviD 1.0.1 and had same problem. :(

[aketon]

One good and one bad thing about ffdshow 20040808!

The good thing:
h.264 decoder can finally play encoded videos with all the analyzer flags enabled with no problems! :)

The bad thing:
Theora is broken!!!:(

[timeismoney]

I feel 0806 is better than 0808

I use 0808 to decode any AVI/MKV encoded by DivX Pro 5.11 will cause any player crash on my system, only videos encoded by DivX Pro 5.11, other version of divx is just fine

BTW, the audio codec is perfect under MPC and any version of WMP below 9.0, but under WMP9 it says like "Can't connect PINs", both AC3/DTS/AAC, and AAC is not friendly to some .aac and media contains aac audio track, m4a is all OK

[arno]

Originally posted by timeismoney
I feel 0806 is better than 0808

I use 0808 to decode any AVI/MKV encoded by DivX Pro 5.11 will cause any player crash on my system, only videos encoded by DivX Pro 5.11, other version of divx is just fine

BTW, the audio codec is perfect under MPC and any version of WMP below 9.0, but under WMP9 it says like "Can't connect PINs", both AC3/DTS/AAC, and AAC is not friendly to some .aac and media contains aac audio track, m4a is all OK

Yep, I agree. And 0808 also seems to hang my BSplayer a lot more often than 0806...

[LigH]

I just tried to run a test with makeAVIS, and read from another user that he does not get filtered output when he uses ffdshow's VfW interface to decode YV12 (he wanted to compare MPEG2Dec3's PP with ffdshow's PP). So I tested three configurations: ffdshow as VfW converter for YV12, ffdshow as frameserver for AVIS fake AVIs, and ffdshow as true decoder for FFV1 AVIs.

Result: No filter (deblocking/deringing, brightness/gamma, noise/grain) was applied to the decoded video in either case.

Is this intentional? Then I'm confused because the VfW interface looks like it provides image filtering, just like the DirectShow filter. Or may it be possible to miss one simple checkbox to enable image filtering in the VfW interface in general?

[marcellus]

Hi, I have two issues to bring to attention:

1. There is a problem with the Logoaway function. In mode "XY" and with a blur value equal and greater than 1 a green shade is cast over the logoaway rectangle that makes it easy to spot on an uniform background (otherwise it wouldnt be seen). The amplitude of this green shading increases in direct proportion with the blur setting (at 0 the problem doesn't show, at 10 is obvious). I use this function in realtime capture to increase compresibility.

Here is a frame that shows the issue. Look at the arrow to see greenish rectangle that should have the same colour as the surounding background.
Logoaway issue example (http://www.marcellusvcd.netfirms.com/Logoaway_01.html)

2. It seems there is an internal error in converting interlaced YUY2 to YV12. When I capture from YUY2 TV and I deinterlace in real time with almost any deinterlacer provided (including dscaler dll's) chroma of the resulted movie is somehow delayed half a frame from luma (This is easy seen at a scene change). When I check the "swap fields" box chroma is behind luma with half a frame. It's like the chroma of 2 consecutive frames is blended together. The only deintelacing method that does not show this "chroma delay" issue is "5 tap low pass" but that is normal since with this method the result is blended frames anyway.

Here is a succesion of frames that show the bug:
Frame1 (http://www.marcellusvcd.netfirms.com/Chr_blend_01.html)
Frame2 (http://www.marcellusvcd.netfirms.com/Chr_blend_02.html)
(Look at the arrow to see the chroma coresponding to the face of the character in frame number 1. The face of the curent frame character is pale because it doesn't have proper chroma like in Frame3.)
Frame3 (http://www.marcellusvcd.netfirms.com/Chr_blend_03.html)

best regards
marcellus

[stephanV]

Some issues with ffdshow's vfw MPEG4 encoder:

1. Enabling b-frames causes files to be undersized by a ratio of 1:5 or more

2. B-frames in combination with qpel crash VirtualDub (several versions) and also graphedit. Here's a crashinfo.txt (http://www.fesite.nl/grotesteph/crashinfo.txt)

Using qpel and b-frames with avs2avi gives me a ICSeqCompressFrame failed: The system cannot find the file specified

I cant remember this behaviour from earlier builds, although im not sure about the b-frames undersizing problems.

hope this is useful. :)

[Bogalvator]

I remember reading somewhere (here?) that ffdshow MPEG-4 with B-frames results in all B-frames being encoded at quant 31 when used in 2 pass - can someone confirm?

The avs2avi could be caused by the fact that ffdshow no longer installs the "ffvfw.dll" for the vfw section (guess work)


On a separate note: I notice that KernelDeint has succesfully been introduced into ffdshow. Would it also be possible to have DGBob integrated in to the deinterlace section without too much difficulty? This would be useful to those who store interlaced encodes on their PC (for things like sports clips, which aren't smooth enough when deinterlaced)

[stephanV]

just to be clear: if i didnt enable b-frames and qpel at the same time, avs2avi did work.

[Andy2222]

Originally posted by Bogalvator
On a separate note: I notice that KernelDeint has succesfully been introduced into ffdshow. Would it also be possible to have DGBob integrated in to the deinterlace section without too much difficulty? This would be useful to those who store interlaced encodes on their PC (for things like sports clips, which aren't smooth enough when deinterlaced)

DGBob was added by milan some days ago and can be used in a new release.

[LigH]

@ milan, athos, Andy2222:

Does ffdshow's VfW interface apply image filters at all? (^)

[Soulhunter]

Source...

http://img69.exs.cx/img69/2770/Source.jpg


Source -> VFW -> Blur 255 -> FFV1

http://img69.exs.cx/img69/1444/Blur255.jpg


Bye

[LigH]

When used as encoder: A little bit, it seems to me (shouldn't 255 soften a lot more?)...

But I'm rather interested in "when used as VfW decoder for AVIS".

[bond]

Originally posted by aketon
The good thing:
h.264 decoder can finally play encoded videos with all the analyzer flags enabled with no problems! :)what analyser flags?

The bad thing: Theora is broken!!!:/

[aketon]

Originally posted by bond
what analyser flags?

:/


Previous releases of ffdshow couldn't play properly the videos encoded with "Inter Analyzer Flag PSUB8x8" and "CABAC" ON at the same time! Now it can play them with no problems! Mplayer can play them also! VLC media player still can't play them properly!


Every time I am trying to encode with THEORA using ffdshow 20040808, the video I get is full of garbage in it! When I swithed back to previous releases of ffdshow, I could encode with no problems!


AS USUAL: "Sorry for my english"


BYE!!!

[bond]

ah yes indeed, afaik this has nothing to do with "analysing" but is basically the block size.
it should have worked for some time already, cause basically the ffmpeg guys fixed this because i made a bugreport at sourceforge, which was two months ago or so
i think i updated my h.264 sticky already with this info, i will have a look, ffmpegs h.264 decoder should be fully useable already i think

[athos]

New release up on SF. I decided not to include skl_drv_mpg.dll. Get it from earlier release, or here:
http://athos.leffe.dnsalias.com/skl_drv_mpg.zip

libavcodec.dll is smaller because i compiled without -funroll-loops.

mpeg-1/mpeg-2 decoding through libmpeg2 seems to work fine in this release, including seeking.

[Gooop]

thanks athos for the new build
xvid with bframes / qpel (don't know what caused the bug actually, didn't have time enough to test) works again, at least for me.

[athos]

Originally posted by Gooop
thanks athos for the new build
xvid with bframes / qpel (don't know what caused the bug actually, didn't have time enough to test) works again, at least for me.
It might have had something to do with the compiler flags i used for GCC. I tried to use only safe flags for this one.

[Worf]

Hi Athos,

i can't see the ffdshow-20040828 files in http://athos.leffe.dnsalias.com/?M=D
Im do something wrong?

Thanks